diff --git a/.env.example b/.env.example
deleted file mode 100644
index 46e07193..00000000
--- a/.env.example
+++ /dev/null
@@ -1,24 +0,0 @@
-# ============================================================================
-# Flask 会话密钥（必须配置）
-# ============================================================================
-SECRET_KEY=your-random-secret-key-change-this
-
-# ============================================================================
-# LangSmith 配置（可选，用于调试追踪）
-# ============================================================================
-LANGSMITH_TRACING=false
-# LANGSMITH_ENDPOINT=https://api.smith.langchain.com
-# LANGSMITH_API_KEY=你的_langsmith_api_key
-# LANGSMITH_PROJECT="error-correction"
-
-# ============================================================================
-# 输出目录配置（可选）
-# ============================================================================
-# 目录配置由 config.py 集中管理（默认：backend/runtime_data/）
-# 如需覆盖运行时输出的根目录，请设置 APP_RUNTIME_DIR（绝对路径）
-# APP_RUNTIME_DIR=
-
-# ============================================================================
-# 注意：API Provider 配置（OpenAI / Anthropic / PaddleOCR）已迁移到数据库，
-# 每个用户在系统设置页面独立管理自己的 API Key 和供应商配置。
-# ============================================================================
diff --git a/.gitattributes b/.gitattributes
new file mode 100644
index 00000000..b763b395
--- /dev/null
+++ b/.gitattributes
@@ -0,0 +1 @@
+backend/models/weight/*.pth filter=lfs diff=lfs merge=lfs -text
diff --git a/.gitignore b/.gitignore
index 513c3f4a..53d02f3d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -36,4 +36,6 @@ dataset/
 
 # 环境变量
 .env
+backend/.env
 PR.md
+TODO.md
diff --git a/CLAUDE.md b/CLAUDE.md
index fc92d1df..9f6f3310 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -19,10 +19,12 @@ cd frontend && npm run dev               # Vite dev server, proxy /api → :5001
 ### 生产构建
 
 ```bash
-cd frontend && npm run build             # 输出到 backend 静态目录
-cd backend && python web_app.py          # Flask 托管前端产物，访问 :5001
+cd frontend && npm run build             # 构建前端产物（目前仅用于部署，Flask 不再托管）
+cd backend && python web_app.py          # 启动纯 API 服务，仅监听 :5001
 ```
 
+> **注意**：`web_app.py` 已重构为纯 API 服务器，不提供前端页面。前端必须通过 Vite dev server（`:5173`）或独立部署访问，直接访问 `:5001` 只会得到 JSON 404。
+
 ### 测试
 
 ```bash
@@ -47,7 +49,7 @@ cd frontend && npm install               # 前端（Node 18+）
 
 ### 环境配置
 
-复制 `.env.example` → `.env`，至少配置一组 LLM API key（DeepSeek 或 ERNIE）和 PaddleOCR token。详见 `.env.example` 注释。
+复制 `backend/.env.example` → 项目根目录 `.env`，必须配置 `SECRET_KEY`。`core/config.py` 从项目根目录读取 `.env`。LLM API Provider 配置（OpenAI / Anthropic / PaddleOCR）已迁移到数据库，用户在系统设置页面管理。SMTP 邮件配置（注册验证码、找回密码）通过 `.env` 的 `APP_SMTP_*` 变量管理。
 
 ---
 
@@ -60,11 +62,13 @@ cd frontend && npm install               # 前端（Node 18+）
 ```
 用户上传文件
   → prepare_input（PDF 转图片）
-  → PaddleOCR API（异步任务，asyncio.gather 并行）
+  → [可选] EnsExam 擦除手写字迹（/api/erase，独立预处理步骤）
+  → PaddleOCR API（/api/ocr，异步并行，返回 bbox 预览）
   → simplify_ocr_results()（block_label 归一化，唯一转换点）
   → split_batch（2 页/批滑动窗口，ThreadPoolExecutor 并行，上限 3）
   → LLM Agent 分割题目（invoke_split）
   → _dedup_questions()（跨批次去重）
+  → _normalize_image_paths()（修复 LLM 篡改的图片路径）
   → LLM Agent 纠错（invoke_correction）
   → 导出 / 保存数据库
 ```
@@ -72,23 +76,33 @@ cd frontend && npm install               # 前端（Node 18+）
 ### 关键模块关系
 
 - **`backend/src/workflow.py`** — LangGraph StateGraph 主工作流，串联所有处理步骤
-- **`backend/error_correction_agent/`** — 题目分割 + OCR 纠错 Agent（`create_agent` 工厂函数）
-- **`backend/solve_agent/`** — 独立的解题 Agent
-- **`backend/teach_agent/`** — 教学讲解 Agent，用于 AI 分析功能
-- **`backend/web_app.py`** — Flask 主应用，路由 + 全局会话状态（`session_lock` 保护）
-- **`backend/config.py`** — 所有运行时路径集中管理，`ensure_dirs()` 显式初始化
-- **`backend/llm.py`** — `init_model()` 统一 LLM 初始化，支持 deepseek / ernie
-- **`backend/db/`** — SQLite + SQLAlchemy ORM，错题库持久化
-- **`frontend/src/App.vue`** — 根组件，全局状态集中于此（无 Pinia/Vuex）
+- **`backend/agents/error_correction/`** — 题目分割 + OCR 纠错 Agent（`create_inner_split_agent` / `create_inner_correct_agent`）
+- **`backend/agents/solve/`** — 独立的解题 Agent（`invoke_solve`）
+- **`backend/agents/teach/`** — 教学讲解 Agent，流式多轮对话（`stream_teach`）
+- **`backend/agents/note/`** — 笔记整理 Agent，OCR → LLM 结构化（`invoke_note_organize`）
+- **`backend/web_app.py`** — Flask 应用工厂，注册 7 个 Blueprint，全局 session 状态在 `state.py`
+- **`backend/routes/`** — 7 个 Blueprint 模块（upload、questions、chat、stats、auth、settings、notes）
+- **`backend/core/state.py`** — 全局会话状态（`session_files`、`session_lock`）
+- **`backend/core/mail.py`** — SMTP 邮件发送（注册验证码、找回密码）
+- **`backend/core/config.py`** — 所有运行时路径 + LLM provider 注册集中管理，`ensure_dirs()` 显式初始化
+- **`backend/core/llm.py`** — `init_model()` 统一 LLM 初始化，支持多 provider
+- **`backend/db/`** — SQLite + SQLAlchemy ORM，错题库持久化；`db/crud/providers.py` 管理用户存储的 API key
 
 ### Flask 路由
 
-- `GET /` → 介绍页 | `GET /app` → Vue 工作台
-- `POST /api/upload` / `/api/split` / `/api/export` / `/api/cancel_file` — 核心工作流 API
-- `GET /api/status` — 系统状态（OCR 配置、可用模型列表）
-- `/api/error-bank` / `/api/subjects` / `/api/question-types` / `/api/stats` — 错题库 CRUD + 统计
-- `/api/chat` / `/api/chat/<id>/messages` / `/api/chat/<id>/stream` — AI 对话（SSE 流式）
-- `/api/ai-analysis` — AI 分析（teach_agent）
+路由实现在 `backend/routes/` 的 7 个 Blueprint 模块中，`web_app.py` 仅做注册：
+
+- `POST /api/upload` / `/api/erase` / `/api/ocr` / `/api/split` / `/api/export` / `/api/cancel_file` — 核心工作流 API（upload.py）
+- `GET /api/image/<filename>` — 图片资源访问（upload.py）
+- `GET /api/status` — 系统状态（OCR 配置、可用模型列表）（upload.py）
+- `GET /api/split-records` — 分割历史记录（upload.py）
+- `/api/error-bank` / `/api/subjects` / `/api/question-types` — 错题库 CRUD（questions.py）
+- `/api/stats` — 统计数据（stats.py）
+- `/api/chat` / `/api/chat/<id>/messages` / `/api/chat/<id>/stream` — AI 对话（SSE 流式）（chat.py）
+- `/api/ai-analysis` — AI 分析（teach_agent）（chat.py）
+- `/api/auth` / `/api/auth/send-code` / `/api/auth/reset-password` — 认证 + 邮箱验证码（auth.py）
+- `/api/settings` — LLM provider 配置（settings.py）
+- `/api/notes` — 笔记 CRUD（notes.py）
 
 ---
 
@@ -107,63 +121,95 @@ cd frontend && npm install               # 前端（Node 18+）
 ### 技术栈
 
 - Vue 3 + `<script setup>`（Composition API，不使用 Options API）
-- Vite 多页面入口：`index.html`（介绍页）+ `app.html`（工作台）
-- Tailwind CSS 3（utility-first，class-based dark mode）
-- HeadlessUI Vue 用于可访问交互组件
+- Vite 单页应用，入口 `app.html`，Vue Router 4 客户端路由
+- Tailwind CSS 3（utility-first，class-based dark mode）+ `@tailwindcss/typography`
+- HeadlessUI Vue 用于可访问交互组件（Listbox、Dialog 等）
 - DOMPurify 用于 HTML 净化（白名单在 `utils.js` 的 `ALLOWED_HTML_TAGS`）
 - 纯 JavaScript（不使用 TypeScript），原生 fetch（不使用 axios），文件上传用 XHR
-- CDN 外部依赖（非 npm）：Font Awesome 6.5、MathJax 3、SortableJS 1.15、Chart.js 4
-
-### 布局
-
-- **PC 端**：左侧固定侧边栏（`aside w-64`）+ 右侧主内容区
-- **移动端**：底部 Tab 导航栏（`fixed bottom-0`）+ 全宽内容区
-- `currentView` ref 控制 `'workspace'` / `'dashboard'` 视图（`v-show` 切换）
+- CDN 外部依赖（非 npm）：Font Awesome 6.5、MathJax 3、SortableJS 1.15、Chart.js 4、marked 15
 
-### 间距节奏（8pt Grid）
+### 目录结构
 
-**所有 padding / margin / gap 只允许使用 4 的倍数**，对应 Tailwind 档位：
-
-| 用途 | 允许值 |
-|------|--------|
-| 微间距（图标、标签内边距） | `p-1` `p-2` `gap-1` `gap-2` |
-| 小间距（按钮、行内元素） | `p-3` `p-4` `gap-3` `gap-4` |
-| 中间距（卡片内边距、区块间距） | `p-6` `p-8` `gap-6` `gap-8` |
-| 大间距（页面边距、区域分割） | `p-10` `p-12` `gap-10` `gap-12` |
-
-禁止使用 `p-5` `p-7` `p-9` `gap-5` `gap-7` 等非 4 倍数值，禁止任意 `px-[17px]` 之类魔法数字。
+```
+frontend/src/
+├── components/           # 可复用组件（55+）
+│   ├── auth/            # 认证：AuthLayout、LoginView、RegisterView、ForgotPasswordModal
+│   └── home/            # 首页：HomeNav、HomeHero、HomeFeatures 等
+├── composables/          # Vue 3 组合式函数（12 个）
+│   ├── useAuth.js       # 认证状态（currentUser、authChecked）
+│   ├── useTheme.js      # 主题切换（isDark、toggleTheme + View Transition 动画）
+│   ├── usePageTransition.js  # 全局页面过渡遮罩
+│   ├── useClickOutside.js    # 点击外部关闭
+│   ├── useImageModal.js      # 图片预览弹窗
+│   ├── useWorkspaceToast.js  # Toast 通知
+│   ├── useSystemStatus.js    # 系统状态（OCR/模型配置）
+│   ├── useSidebarIndicator.js # 侧边栏 active 指示器动画
+│   ├── useAiChatSessions.js  # AI 对话会话管理
+│   ├── useFileUpload.js      # 文件上传状态
+│   ├── useSplitPipeline.js   # 分割流水线（擦除→OCR→分割）
+│   └── useQuestionList.js    # 题目列表状态
+├── router/index.js       # 路由配置
+├── views/                # 页面级组件
+│   ├── HomeView.vue     # 首页（强制暗色主题）
+│   └── WorkspaceView.vue # 工作台主容器（侧边栏 + 内容区）
+├── api.js                # 集中式 API 层（50+ 函数）
+├── utils.js              # 工具函数（sanitizeHtml、renderMarkdown、typesetMath）
+├── style.css             # 全局样式 + Tailwind 自定义类
+├── App.vue               # 根组件
+└── main.js               # Vue 应用入口
+```
 
-### 排版层级（Typography）
+### 路由
 
-字号只允许以下 7 档，**禁止使用其他值**（包括 `text-[13px]`、`text-[11px]` 等任意值）：
+| 路径 | 组件 | 说明 |
+|------|------|------|
+| `/` | HomeView | 首页 |
+| `/auth/login` | LoginView | 登录 |
+| `/auth/register` | RegisterView | 注册 |
+| `/app/workspace` | WorkspaceView | 录入工作台 |
+| `/app/dashboard` | Dashboard | 数据统计 |
+| `/app/error-bank` | ErrorBank | 错题库 |
+| `/app/notes` | NoteView | 笔记整理 |
+| `/app/ai-chat` | ChatPage | AI 对话 |
+| `/app/review` | ReviewView | 复习计划 |
+| `/app/settings` | SettingsView | 系统设置 |
+| `/app/split-history` | SplitHistory | 分割历史 |
+| `/app/chat` | ChatView | 独立对话 |
+
+路由 meta 字段：`layout`（`'home'` / `'auth'` / `'app'`）、`requiresAuth`（boolean）
 
-| 层级 | Tailwind | 像素 | 行高 | 用途 |
-|------|----------|------|------|------|
-| 超大标题 | `text-4xl` | 36px | `leading-tight`（1.25） | 落地页 Hero、各区块主标题 |
-| 大标题 | `text-3xl` | 30px | `leading-tight`（1.25） | 页面主标题、工作台区块标题 |
-| 标题 | `text-2xl` | 24px | `leading-snug`（1.375） | 区块标题、Modal 标题 |
-| 副标题 | `text-xl` | 20px | `leading-snug` | 卡片标题、侧边栏项 |
-| 正文 | `text-base` | 16px | `leading-relaxed`（1.625） | 题目内容、说明文字 |
-| 辅助 | `text-sm` | 14px | `leading-relaxed` | 标签、次要信息 |
-| 标注 | `text-xs` | 12px | `leading-normal`（1.5） | 元数据、时间戳、徽标 |
+### 布局
 
-**信息密度原则**：每张 Card 只允许 1 个主信息 + 2 个辅信息，禁止堆砌超过 3 层文字层级。
+- **PC 端**：左侧固定侧边栏（`aside w-64`）+ 右侧主内容区，导航分组可折叠
+- **移动端**：底部 Tab 导航栏（`fixed bottom-0`）+ 全宽内容区
+- `currentView` ref 控制视图切换，路由参数 `/app/:view?/:subview?`
 
 ### 设计风格
 
-**配色 Token（只允许以下色系，禁止随意引入彩虹色）：**
+**Glass Morphism + Linear 风格**，以暗色为主、亮色兼容。
+
+**配色 Token：**
 
-| 语义 | 亮色 | 暗色 |
+| 语义 | 暗色 | 亮色 |
 |------|------|------|
-| 背景 | `white` / `slate-50` | `slate-900` / `slate-950` / `[#0A0A0F]` |
-| 卡片面 | `white/70` + `backdrop-blur-xl` | `white/[0.03]` + `backdrop-blur-xl` |
-| 边框 | `border-slate-200/60` | `border-white/10` |
-| 主操作 | `blue-600` / `indigo-600` | `indigo-500` |
-| 成功 | `emerald-600` | `emerald-400` |
-| 错误 | `rose-600` | `rose-400` |
-| 中性文字 | `slate-900` / `slate-700` / `slate-500` / `slate-400` | `white` / `slate-200` / `slate-400` / `slate-500` |
+| 页面背景 | `[#0A0A0F]` / `slate-950` | `white` / `slate-50` |
+| 卡片 / 面板 | `white/[0.02]` ~ `white/[0.05]` + `backdrop-blur-xl` | `white/70` + `backdrop-blur-xl` |
+| 边框 | `white/[0.06]` ~ `white/[0.10]` | `slate-200/60` |
+| 品牌色 | `rgb(129,115,223)` ~ `rgb(145,132,235)` | `indigo-600` / `blue-600` |
+| 成功 | `emerald-400` | `emerald-600` |
+| 错误 | `rose-400` | `rose-600` |
+| 主要文字 | `white` / `[#d0d6e0]` | `slate-900` |
+| 次要文字 | `[#8a8f98]` / `slate-400` | `slate-500` / `slate-700` |
+| 占位文字 | `[#62666d]` | `slate-400` |
+
+**Glass 卡片标准写法：**
+```
+bg-white/[0.02] border border-white/[0.06] rounded-xl
+border-t-white/[0.15] border-b-white/[0.03]
+hover:bg-white/[0.04] transition-all
+```
 
-**圆角 Token：**
+**圆角：**
 
 | 元素 | 值 |
 |------|-----|
@@ -172,35 +218,120 @@ cd frontend && npm install               # 前端（Node 18+）
 | 标签、Badge、Pill | `rounded-full` |
 | 图标容器、小元素 | `rounded-lg` |
 
-**阴影层级（只允许 2 层，禁止滥用大投影）：**
+**阴影（只允许 2 层）：**
 
 | 状态 | 值 |
 |------|-----|
 | 默认 | `shadow-sm` |
 | Hover / 激活 | `shadow-md` 或品牌色光晕 `shadow-blue-500/20` |
 
-- **所有元素必须包含 `dark:` 变体**
-- 主题切换支持 View Transitions API 圆形扩散动画
-
-### 按钮样式
+### 按钮
 
 | 类型 | 样式 |
 |------|------|
-| 主按钮 | `h-10 rounded-xl bg-blue-600 text-white text-sm font-bold` + hover 光晕 |
-| 成功按钮 | `h-10 rounded-xl border border-emerald-500/30 bg-emerald-50/80 text-emerald-700 text-sm font-bold` |
-| 次按钮 | `h-10 rounded-xl border border-slate-200/60 bg-white/60 text-slate-700 text-sm font-bold` |
-| 通用属性 | `inline-flex items-center justify-center gap-2 transition-all` |
+| 品牌按钮 | `brand-btn`（style.css 中定义，frosted glass 风格） |
+| 主按钮 | `btn-primary`（蓝→靛渐变，shadow-blue-500/20） |
+| 成功按钮 | `btn-success`（翠绿渐变） |
+| 次按钮 | `btn-secondary`（边框 + 透明底） |
+| 通用属性 | `inline-flex items-center justify-center gap-2 transition-all h-10 rounded-xl text-sm font-bold` |
 | 禁用 | `disabled:cursor-not-allowed disabled:opacity-50` |
 
-按钮高度统一 `h-10`（40px），图标尺寸统一 `size-4`（16px）或 `size-5`（20px）。
+按钮高度统一 `h-10`（40px），图标尺寸 `size-4`（16px）或 `size-5`（20px）。
+
+### 筛选器
+
+使用 `filter-pill` / `filter-pill--active` 类（style.css），激活态为品牌紫色 `rgb(145,132,235)`。
+
+### 排版
+
+字号只允许以下 7 档：
+
+| 层级 | Tailwind | 用途 |
+|------|----------|------|
+| 超大标题 | `text-4xl` | 落地页 Hero |
+| 大标题 | `text-3xl` | 页面主标题 |
+| 标题 | `text-2xl` | 区块标题、Modal 标题 |
+| 副标题 | `text-xl` | 卡片标题 |
+| 正文 | `text-base` | 题目内容 |
+| 辅助 | `text-sm` | 标签、次要信息 |
+| 标注 | `text-xs` | 元数据、时间戳 |
 
-### 状态管理与代码规范
+禁止 `text-[13px]`、`text-[11px]` 等任意值。
 
-- 无状态管理库，全局状态集中在 `App.vue`（ref/reactive/computed/watch）
+### 间距（8pt Grid）
+
+只允许 4 的倍数：`p-1` `p-2` `p-3` `p-4` `p-6` `p-8` `p-10` `p-12`。禁止 `p-5` `p-7` `p-9` 和 `px-[17px]` 等魔法数字。
+
+### 主题系统
+
+- `useTheme` composable 管理暗色/亮色切换
+- `document.documentElement.classList.toggle('dark')` + `localStorage.theme`
+- 切换动画：View Transition API 圆形扩散（`clip-path` 圆形展开）
+- `app.html` 内有防白闪脚本（DOM 渲染前读取 localStorage 设置 `dark` 类）
+- **所有新增元素必须包含 `dark:` 变体**
+- 首页强制暗色主题，离开时恢复用户原主题
+
+### 状态管理
+
+- 无 Vuex/Pinia，轻量 ref 方案
+- 全局认证状态：`useAuth` composable（`currentUser`、`authChecked`）
+- 页面级状态：集中在 `WorkspaceView.vue`（ref/reactive/computed/watch）
 - 父子通信：props 向下，emits 向上
+- 跨组件共享通过 composable 单例 ref
+
+### API 调用
+
+- 集中在 `api.js`，所有函数导出供组件调用
+- 统一错误处理：检查 `resp.ok` + `data.success`，throw 描述性错误
+- 模型请求用 `_buildModelBody(modelProvider, modelName, extra)` 构建
+- SSE 流式：`streamChat()` 支持 abort signal
+- 文件上传用 XHR（支持 progress 回调）
+
+### 内容渲染
+
+- `renderMarkdown(text)` — marked.js 解析 + DOMPurify 净化
+- `sanitizeHtml(html)` — 白名单标签净化（table、img 等）
+- `typesetMath(el)` — MathJax 公式渲染（行内 `$...$`、独占行 `$$...$$`）
+- 渲染后必须调用 `typesetMath` 触发公式排版
+
+### 图标
+
+- 主要：Font Awesome `<i class="fas fa-icon-name"></i>`
+- 备用：lucide-vue-next（按需导入）
+- 加载中：`fa-spinner fa-spin`
+
+### 组件开发规范
+
+```vue
+<script setup>
+import { ref, computed, watch, onMounted } from 'vue'
+
+const props = defineProps({ ... })
+const emit = defineEmits(['event-name'])
+
+// 状态、计算属性、生命周期...
+</script>
+
+<template>
+  <!-- 模板 -->
+</template>
+
+<style scoped>
+/* 仅组件私有样式，全局样式放 style.css */
+</style>
+```
+
 - UI 文案使用中文，API 路径前缀 `/api/`
 - 新功能提取为独立 `.vue` 组件放 `src/components/`
+- 认证组件放 `components/auth/`，首页组件放 `components/home/`
 - 关键 DOM 元素添加语义类名或 `data-testid`
+- Modal/Dropdown 使用 `useClickOutside` 处理外部点击关闭
+
+### Vite 配置
+
+- API 代理：`/api`、`/images`、`/download`、`/erased`、`/uploads` → `http://localhost:5001`
+- SPA 路由：`/auth*`、`/app*` → `app.html`
+- 构建产物：`dist/`
 
 ---
 
@@ -209,7 +340,7 @@ cd frontend && npm install               # 前端（Node 18+）
 ### 模块级副作用
 
 - **禁止在模块顶层执行有副作用的操作**（创建目录、写文件、启动连接等）
-- `config.py` 目录创建通过 `ensure_dirs()` 显式调用
+- `core/config.py` 目录创建通过 `ensure_dirs()` 显式调用
 - `load_dotenv()` 仅在入口文件或需要环境变量的模块中调用
 
 ### 数据库
@@ -220,7 +351,7 @@ cd frontend && npm install               # 前端（Node 18+）
 
 ### 线程安全
 
-- Flask 全局状态修改必须在 `session_lock` 保护下
+- 全局会话状态集中在 `core/state.py`，修改必须在 `session_lock` 保护下
 - 锁内用原地修改（`.remove()`, `.append()`），**不要用推导式重新赋值**
 - Agent 缓存通过 `_agent_cache_lock` 保护并发初始化
 
@@ -228,9 +359,9 @@ cd frontend && npm install               # 前端（Node 18+）
 
 - 结构化输出优先用 `create_agent` + `ToolStrategy`
 - 已知问题：`ToolStrategy` 与 DeepSeek 的 `handle_errors` 重试不兼容，大数据量可能触发 400
-- ernie 用 `model.with_structured_output()`（不支持 function calling）
-- 轻量任务用 `DEEPSEEK_LIGHT_MODEL_NAME` / `ERNIE_LIGHT_MODEL_NAME` 小模型
+- 不支持 function calling 的模型用 `model.with_structured_output()`
 - `invoke_split` / `invoke_correction` 是统一调用入口，屏蔽 provider 差异
+- API Provider 配置（key、模型名）存储在数据库中，用户通过系统设置页面管理
 
 ### OCR 数据处理
 
@@ -240,14 +371,18 @@ cd frontend && npm install               # 前端（Node 18+）
 
 ### 文件与路径
 
-- 所有运行时路径通过 `config.py` 集中管理，禁止硬编码
+- 所有运行时路径通过 `core/config.py` 集中管理，禁止硬编码
 - 文件名解析用 `os.path.splitext()`，**不要用 `rsplit('.', 1)`**
 - 文件上传用 `uuid.uuid4().hex` 生成安全文件名
 
 ### 环境变量
 
-- `.env` 不入版本控制，`.env.example` 作为模板
-- 新增配置项必须同步更新 `.env.example`
+- `.env` 放项目根目录，不入版本控制；`backend/.env.example` 作为模板
+- `core/config.py` 的 `Settings` 使用 `env_prefix="APP_"`，从项目根目录 `.env` 读取
+- LLM API Provider 配置已迁移到数据库，不再通过 `.env` 管理
+- Flask 级配置（`SECRET_KEY`、`FLASK_DEBUG`、`LANGSMITH_*`）仍通过 `.env` 管理
+- SMTP 邮件配置通过 `APP_SMTP_*` 变量管理（`core/mail.py`）
+- 新增 `.env` 配置项必须同步更新 `backend/.env.example`
 - 可选项用 `os.getenv("KEY")` + 代码中提供默认值
 
 ---
diff --git a/README.md b/README.md
index 66e80ad5..e0b214e7 100644
--- a/README.md
+++ b/README.md
@@ -1,40 +1,59 @@
-# 错题本生成系统
+# 智卷 — 智能错题本生成系统
 
-基于 PaddleOCR + LangChain Agent 的智能错题本生成系统。上传试卷 PDF 或图片，自动识别文档结构、智能分割题目、OCR 纠错，导出为 Markdown 错题本并存入错题库，支持 AI 解题与教学讲解。
+基于 PaddleOCR + LangChain Agent 的智能错题本生成系统。上传试卷 PDF 或图片，自动识别文档结构、智能分割题目、OCR 纠错，导出为 Markdown 错题本并存入错题库。支持 AI 解题与教学讲解、笔记整理、独立 AI 对话、错题复习计划与数据统计。
 
 ## 功能
 
-- **智能分割**：LLM Agent 自动识别题目边界，支持批次并行处理
+- **智能分割**：LLM Agent 自动识别题目边界，支持批次并行处理，跨页去重
 - **OCR 纠错**：LLM 对识别结果进行结构化纠错，还原原题格式
-- **错题库**：SQLite 持久化存储，支持按科目、题型、日期筛选和统计
+- **手写擦除**：EnsExam 模型自动擦除手写字迹，提升 OCR 精度
+- **OCR 预览**：识别结果 bbox 框标注叠加在原图上，直观预览
+- **错题库**：SQLite 持久化存储，支持按科目、题型、知识点筛选和统计
+- **笔记整理**：上传手写笔记图片，OCR + LLM 自动结构化整理为 Markdown
 - **AI 解题**：独立解题 Agent，逐步给出解题过程
-- **AI 分析**：教学讲解 Agent，深度分析题目知识点
-- **AI 对话**：多轮对话，支持 SSE 流式输出
-- **导出**：导出为 Markdown 文件
+- **AI 教学**：教学讲解 Agent，绑定错题上下文的一对一辅导
+- **AI 对话**：独立多轮对话，支持 SSE 流式输出和深度思考（DeepSeek Reasoner）
+- **复习计划**：间隔复习，追踪错题掌握状态
+- **数据统计**：Dashboard 可视化，按科目/题型/时间维度分析
+- **导出**：导出为 Markdown 文件，按大题分组
 
 ## 项目结构
 
 ```
-├── backend/                         # Flask 后端（API + 静态资源托管）
-│   ├── config.py                    # 集中配置（路径、目录）
-│   ├── llm.py                       # LLM 初始化（DeepSeek / 文心）
-│   ├── web_app.py                   # Flask 主应用（路由 + 会话状态）
+├── backend/                         # Flask 后端（纯 API 服务）
+│   ├── core/                        # 核心模块
+│   │   ├── config.py                # 集中配置（路径、Settings）
+│   │   ├── llm.py                   # LLM 初始化（多 provider 支持）
+│   │   ├── state.py                 # 全局会话状态（session_files、锁）
+│   │   └── mail.py                  # SMTP 邮件发送
+│   ├── web_app.py                   # Flask 应用工厂 + Blueprint 注册
+│   ├── routes/                      # 7 个 Blueprint（upload、questions、chat、stats、auth、settings、notes）
 │   ├── src/                         # 核心模块（LangGraph workflow、OCR 客户端、工具函数）
-│   ├── error_correction_agent/      # LangChain Agent（题目分割、OCR 纠错、科目识别）
-│   ├── solve_agent/                 # 解题 Agent
-│   ├── teach_agent/                 # 教学讲解 Agent（AI 分析功能）
-│   ├── benchmark/                   # 模型评测（数据采集、评测运行、指标计算）
-│   ├── db/                          # SQLite 数据库（ORM 模型、CRUD）
-│   └── tests/                       # 后端测试（19 个测试模块）
-├── frontend/                        # Vue 3 + Vite + Tailwind CSS 前端
-│   ├── index.html                   # 介绍落地页
-│   ├── app.html                     # Vue 工作台入口
+│   ├── agents/                      # LangChain Agent
+│   │   ├── error_correction/        # 题目分割 + OCR 纠错
+│   │   ├── solve/                   # 解题 Agent
+│   │   ├── teach/                   # 教学讲解 Agent
+│   │   └── note/                    # 笔记整理 Agent
+│   ├── db/                          # SQLite + SQLAlchemy ORM
+│   │   ├── models.py               # 数据模型（User、Question、Note、Chat 等）
+│   │   ├── crud/                    # CRUD 模块化包
+│   │   └── migrate.py              # 数据库自动迁移
+│   ├── benchmark/                   # 模型评测
+│   └── tests/                       # 后端测试
+├── frontend/                        # Vue 3 + Vite + Tailwind CSS
+│   ├── app.html                     # SPA 入口
 │   └── src/
-│       ├── App.vue                  # 根组件（侧边栏布局 + 视图路由）
-│       ├── components/              # 18 个功能组件
+│       ├── views/                   # 页面级组件（HomeView、WorkspaceView）
+│       ├── components/              # 55+ 功能组件
+│       │   ├── auth/               # 认证（登录、注册、找回密码）
+│       │   └── home/               # 首页组件
+│       ├── composables/             # 组合式函数（useAuth、useTheme 等）
+│       ├── router/                  # Vue Router 路由配置
+│       ├── api.js                   # 集中式 API 层
+│       ├── utils.js                 # 工具函数（Markdown 渲染、MathJax、DOMPurify）
 │       └── __tests__/               # 前端测试（Vitest）
-├── example_uploads/                 # 示例测试文件（图片 + PDF）
-├── .env.example                     # 环境变量模板
+├── example_uploads/                 # 示例测试文件
+├── backend/.env.example             # 环境变量模板
 └── requirements.txt                 # Python 依赖
 ```
 
@@ -55,27 +74,15 @@ cd frontend && npm install
 ### 2. 配置环境变量
 
 ```bash
-copy .env.example .env
+# 复制模板到项目根目录
+cp backend/.env.example .env
 ```
 
-编辑 `.env`，填写以下必需项：
+编辑 `.env`，必须配置 `SECRET_KEY`。
 
-```dotenv
-# LLM API（OpenAI 兼容接口，支持 OpenAI / DeepSeek / Qwen / Moonshot 等）
-OPENAI_API_KEY=your_key
-# OPENAI_BASE_URL=https://api.deepseek.com   # 留空则使用 OpenAI 官方；填写后可切换 DeepSeek 等供应商
-# OPENAI_MODEL_NAME=gpt-4o-mini              # 默认模型（可选）
+LLM API Provider（OpenAI / Anthropic / PaddleOCR）配置已迁移到数据库，启动后在系统设置页面管理。
 
-# Anthropic API（可选，与 OpenAI 兼容接口二选一或同时配置）
-# ANTHROPIC_API_KEY=your_key
-
-# PaddleOCR API（文档结构解析，V2 异步任务接口）
-PADDLEOCR_API_URL=https://paddleocr.aistudio-app.com/api/v2/ocr/jobs
-PADDLEOCR_API_TOKEN=your_token
-PADDLEOCR_MODEL=PaddleOCR-VL-1.5
-```
-
-> LLM 至少配置一种（OpenAI 兼容接口或 Anthropic），前端会自动检测可用模型。轻量模型、LangSmith 追踪、PaddleOCR 特性开关等可选配置见 `.env.example`。
+SMTP 邮件配置（注册验证码、找回密码）通过 `.env` 的 `APP_SMTP_*` 变量管理，详见 `backend/.env.example`。
 
 ### 3. 启动
 
@@ -89,19 +96,11 @@ cd backend && python web_app.py
 cd frontend && npm run dev
 ```
 
-前端开发服务器会自动将 `/api`、`/images`、`/download` 请求代理到后端 `localhost:5001`。
-
-**生产模式**（Flask 托管前端静态资源）：
+前端开发服务器会自动将 `/api`、`/images`、`/download`、`/erased`、`/uploads` 请求代理到后端 `localhost:5001`。
 
-```bash
-# 构建前端
-cd frontend && npm run build
-
-# 启动后端（自动托管构建产物）
-cd backend && python web_app.py
-```
+访问 **http://localhost:5173** 即可使用。
 
-访问 **http://localhost:5001** 即可使用。
+> **注意**：后端已重构为纯 API 服务器，不提供前端页面。直接访问 `:5001` 只会得到 JSON 404。
 
 ## 支持的文件格式
 
diff --git a/TODO.md b/TODO.md
deleted file mode 100644
index ca6fc76d..00000000
--- a/TODO.md
+++ /dev/null
@@ -1,12 +0,0 @@
-# TODO
-
-## 用户级 API Key 存储
-
-每个用户独立配置自己的 LLM API Key（AES-256 加密存 DB），替代系统全局 .env 配置。
-
-- [ ] DB：用户表新增加密 api_key 字段（per provider）
-- [ ] 后端：加解密工具函数（AES-256，密钥从 .env 读取）
-- [ ] 后端：设置接口（POST 保存 / DELETE 删除 / GET 状态查询，不返回明文）
-- [ ] 后端：调用 LLM 时优先使用用户 key，fallback 到系统全局 key
-- [ ] 前端：设置页 API Key 管理 UI（输入、保存、删除、状态显示）
-- [ ] .env.example：新增 API_KEY_ENCRYPTION_SECRET 配置项
diff --git a/WEB_APP_GUIDE.md b/WEB_APP_GUIDE.md
deleted file mode 100644
index ea5324fe..00000000
--- a/WEB_APP_GUIDE.md
+++ /dev/null
@@ -1,443 +0,0 @@
-# Web应用使用指南
-
-## 📱 功能简介
-
-Web应用提供了一个可视化界面，用于测试错题本生成系统的完整工作流程。
-
-### 主要功能
-
-1. ✅ **文件上传** - 支持拖拽或点击上传PDF/图片
-2. ✅ **自动OCR解析** - 调用PaddleOCR API进行文档结构化解析
-3. ✅ **AI题目分割** - 使用DeepSeek Agent智能分割题目
-4. ✅ **题目预览** - 可视化展示分割后的题目
-5. ✅ **选择导出** - 勾选需要的题目并导出为Markdown格式
-
----
-
-## 🚀 快速启动
-
-### 1. 确保环境配置完成
-
-检查 `.env` 文件是否包含必要的配置：
-
-```bash
-# 必需配置
-PADDLEOCR_API_URL=https://...
-PADDLEOCR_API_TOKEN=your_token
-DEEPSEEK_API_KEY=your_key
-
-# 可选配置
-LANGSMITH_TRACING=true
-LANGSMITH_API_KEY=your_key
-```
-
-### 2. 启动Web服务
-
-```bash
-python web_app.py
-```
-
-### 3. 访问界面
-
-打开浏览器访问：**http://localhost:5001**
-
----
-
-## 📋 使用流程
-
-### 步骤1: 上传文件
-
-- **方式1**: 点击上传区域选择文件
-- **方式2**: 拖拽文件到上传区域
-
-**支持格式**:
-- PDF: `.pdf`
-- 图片: `.png`, `.jpg`, `.jpeg`, `.bmp`, `.tiff`, `.webp`
-
-**文件大小**: 最大 50MB
-
-### 步骤2: OCR解析
-
-文件上传后，系统会自动：
-1. 将PDF转换为图片（如果是PDF）
-2. 调用PaddleOCR API进行文档结构化解析
-3. 保存OCR结果到 `output/struct/` 目录
-
-**进度显示**:
-- 图片数量
-- OCR解析完成状态
-
-### 步骤3: 分割题目
-
-点击 **"🤖 开始分割题目"** 按钮：
-
-1. 系统调用DeepSeek Agent
-2. Agent分析OCR结果
-3. 智能识别题号、题型、选项
-4. 保存分割结果到 `results/questions.json`
-
-**预计时间**: 10-30秒/页（取决于API响应速度）
-
-### 步骤4: 预览和导出
-
-分割完成后：
-
-1. **预览题目**: 查看所有识别的题目
-2. **选择题目**: 勾选需要导出的题目
-3. **导出错题本**: 点击 **"📥 导出错题本"** 按钮
-4. **下载文件**: 点击下载链接获取Markdown文件
-
----
-
-## 🔧 API端点说明
-
-### 1. 系统状态检查
-
-```http
-GET /api/status
-```
-
-**响应**:
-```json
-{
-  "success": true,
-  "status": {
-    "paddleocr_configured": true,
-    "deepseek_configured": true,
-    "langsmith_enabled": false,
-    "output_dirs": {
-      "pages": "output/pages",
-      "struct": "output/struct",
-      "results": "results"
-    }
-  }
-}
-```
-
-### 2. 文件上传和OCR解析
-
-```http
-POST /api/upload
-Content-Type: multipart/form-data
-
-file: <文件>
-```
-
-**响应**:
-```json
-{
-  "success": true,
-  "message": "文件处理成功",
-  "result": {
-    "image_count": 3,
-    "ocr_count": 3,
-    "image_paths": ["output/pages/test_page_001.png", ...]
-  }
-}
-```
-
-### 3. Agent分割题目
-
-```http
-POST /api/split
-```
-
-**响应**:
-```json
-{
-  "success": true,
-  "message": "成功分割 5 道题目",
-  "questions": [
-    {
-      "question_id": "1",
-      "question_type": "选择题",
-      "content_blocks": [...],
-      "options": ["A. ...", "B. ..."],
-      ...
-    }
-  ]
-}
-```
-
-### 4. 导出错题本
-
-```http
-POST /api/export
-Content-Type: application/json
-
-{
-  "selected_ids": ["1", "2", "3"]
-}
-```
-
-**响应**:
-```json
-{
-  "success": true,
-  "message": "错题本导出成功",
-  "output_path": "results/wrongbook.md"
-}
-```
-
-### 5. 获取题目列表
-
-```http
-GET /api/questions
-```
-
-**响应**:
-```json
-{
-  "success": true,
-  "questions": [...]
-}
-```
-
-### 6. 下载文件
-
-```http
-GET /download/wrongbook.md
-```
-
-直接下载Markdown文件
-
----
-
-## 🎨 界面功能说明
-
-### 系统状态指示器
-
-顶部显示系统配置状态：
-- ✓ **PaddleOCR**: API已配置
-- ✓ **DeepSeek**: API已配置
-- ✓ **LangSmith追踪**: 调试追踪已启用
-
-### 步骤指示器
-
-4步流程可视化：
-1. **上传文件** - 上传PDF或图片
-2. **OCR解析** - 文档结构化识别
-3. **分割题目** - AI智能分割
-4. **预览导出** - 选择和导出
-
-### 题目卡片
-
-每道题目显示：
-- 题号（如"题目 1"）
-- 题型标签（选择题/填空题/解答题）
-- 题目内容（文本、公式、图片引用）
-- 选项（如果是选择题）
-- 复选框（用于选择导出）
-
----
-
-## ⚙️ 配置说明
-
-### 端口配置
-
-默认端口: **5001**
-
-修改端口（在 `web_app.py` 最后一行）:
-```python
-app.run(host='0.0.0.0', port=8080, debug=True)
-```
-
-### 文件大小限制
-
-默认: **50MB**
-
-修改限制（在 `web_app.py` 配置部分）:
-```python
-app.config['MAX_CONTENT_LENGTH'] = 100 * 1024 * 1024  # 100MB
-```
-
-### 上传目录
-
-默认: `uploads/`
-
-修改目录（在 `web_app.py` 配置部分）:
-```python
-UPLOAD_FOLDER = 'my_uploads'
-```
-
----
-
-## 🐛 常见问题
-
-### Q1: 上传文件后长时间没有响应？
-
-**原因**: PaddleOCR API响应较慢或网络问题
-
-**解决**:
-1. 检查网络连接
-2. 查看浏览器控制台的网络请求
-3. 检查 `.env` 中的API配置是否正确
-
-### Q2: Agent分割失败？
-
-**原因**: DeepSeek API配置错误或OCR结果为空
-
-**解决**:
-1. 检查 `DEEPSEEK_API_KEY` 是否正确
-2. 查看 `output/struct/` 目录下是否有OCR结果
-3. 检查 `results/split_issues.jsonl` 查看Agent记录的问题
-
-### Q3: 题目预览显示为空？
-
-**原因**: Agent未能正确识别题目或保存失败
-
-**解决**:
-1. 检查 `results/questions.json` 是否存在
-2. 手动查看JSON文件内容
-3. 启用LangSmith追踪查看Agent执行日志
-
-### Q4: 导出的Markdown文件在哪里？
-
-**位置**: `results/wrongbook.md`
-
-**下载**: 点击页面上的下载链接，或直接访问 `/download/wrongbook.md`
-
----
-
-## 🔍 调试技巧
-
-### 1. 查看系统日志
-
-启动Web应用后，终端会显示所有请求日志：
-```
-127.0.0.1 - - [27/Jan/2026 10:30:15] "POST /api/upload HTTP/1.1" 200 -
-127.0.0.1 - - [27/Jan/2026 10:30:45] "POST /api/split HTTP/1.1" 200 -
-```
-
-### 2. 检查中间结果
-
-**标准化图片**:
-```bash
-ls output/pages/
-# test_page_001.png, test_page_002.png, ...
-```
-
-**OCR结果**:
-```bash
-ls output/struct/
-# test_page_001_struct.json, ...
-```
-
-**题目JSON**:
-```bash
-cat results/questions.json
-```
-
-**问题日志**:
-```bash
-cat results/split_issues.jsonl
-```
-
-### 3. 启用LangSmith追踪
-
-在 `.env` 中设置：
-```bash
-LANGSMITH_TRACING=true
-LANGSMITH_API_KEY=your_key
-LANGSMITH_PROJECT=error-correction-web
-```
-
-然后访问 https://smith.langchain.com 查看Agent执行轨迹。
-
-### 4. 浏览器开发者工具
-
-按 `F12` 打开开发者工具：
-- **Console**: 查看JavaScript错误
-- **Network**: 查看API请求和响应
-- **Application**: 查看本地存储
-
----
-
-## 📊 性能优化
-
-### 1. 减少处理时间
-
-**方法1**: 降低图片DPI
-```python
-# 在 src/utils.py 的 prepare_input 函数中
-images = convert_from_path(file_path, dpi=200)  # 默认300
-```
-
-**方法2**: 关闭部分OCR功能
-```bash
-# .env
-PADDLEOCR_USE_DOC_ORIENTATION=false
-PADDLEOCR_USE_DOC_UNWARPING=false
-PADDLEOCR_USE_CHART_RECOGNITION=false
-```
-
-### 2. 提高并发处理能力
-
-使用 gunicorn 代替 Flask 开发服务器：
-```bash
-pip install gunicorn
-gunicorn -w 4 -b 0.0.0.0:5001 web_app:app
-```
-
-### 3. 添加缓存
-
-对于相同文件，可以缓存OCR结果避免重复调用API。
-
----
-
-## 🔒 安全注意事项
-
-### 1. 文件上传安全
-
-- ✅ 已限制文件扩展名
-- ✅ 已使用 `secure_filename` 处理文件名
-- ✅ 已限制文件大小（50MB）
-
-### 2. API密钥保护
-
-- ⚠️ 不要将 `.env` 文件提交到版本控制
-- ⚠️ 生产环境使用更安全的密钥管理方案
-
-### 3. 部署建议
-
-如果部署到公网：
-- 添加用户认证
-- 使用HTTPS
-- 限制请求频率
-- 定期清理临时文件
-
----
-
-## 🎯 扩展功能建议
-
-### 1. 批量处理
-
-支持一次上传多个文件并批量处理
-
-### 2. 历史记录
-
-保存处理历史，支持查看和重新导出
-
-### 3. 自定义模板
-
-允许用户自定义错题本的Markdown模板
-
-### 4. 在线编辑
-
-在导出前支持在线编辑题目内容
-
----
-
-## 📞 技术支持
-
-如遇到问题：
-
-1. 查看本文档的"常见问题"部分
-2. 检查 `docs/PROJECT_STRUCTURE.md` 了解系统架构
-3. 查看终端日志和浏览器控制台
-4. 启用LangSmith追踪进行调试
-
----
-
-**文档版本**: v1.0.0
-**最后更新**: 2026-01-27
diff --git a/backend/.env.example b/backend/.env.example
new file mode 100644
index 00000000..0bf1ebb6
--- /dev/null
+++ b/backend/.env.example
@@ -0,0 +1,70 @@
+# ============================================================================
+# Flask API 服务配置 SECRET_KEY必须配置!
+# ============================================================================
+SECRET_KEY=your-random-secret-key-change-this
+# 开发模式开关（true 启用热重载和调试信息，生产环境务必设为 false）
+FLASK_DEBUG=false
+
+# ============================================================================
+# LangSmith 配置（可选，用于调试追踪）
+# ============================================================================
+LANGSMITH_TRACING=false
+# LANGSMITH_ENDPOINT=https://api.smith.langchain.com
+# LANGSMITH_API_KEY=你的_langsmith_api_key
+# LANGSMITH_PROJECT="error-correction"
+
+# ============================================================================
+# 输出目录配置（可选）
+# ============================================================================
+# 目录配置由 config.py 集中管理（默认：backend/runtime_data/）
+# 如需覆盖运行时输出的根目录，请设置 APP_RUNTIME_DIR（绝对路径）
+# APP_RUNTIME_DIR=
+
+# ============================================================================
+# 文字擦除模型权重（可选）
+# ============================================================================
+# 默认路径：backend/models/weight/best.pth
+# 如需指定其他位置，取消注释并填写绝对路径
+# APP_MODEL_PATH=/path/to/your/best.pth
+
+# ============================================================================
+
+# ============================================================================
+# 验证码邮件 — SMTP 配置（注册验证码 + 找回密码共用）
+#
+# 【必填项】APP_SMTP_HOST / APP_SMTP_USER / APP_SMTP_PASSWORD / APP_SMTP_FROM
+#   未填写时无法发送验证码邮件。
+#   开发调试时可设置 FLASK_DEBUG=true，验证码会打印到后端日志，无需真实 SMTP。
+#
+# 【发件账号说明】
+#   APP_SMTP_USER  — 发件邮箱地址（同时作为 SMTP 登录用户名）
+#   APP_SMTP_FROM  — 邮件显示的发件人地址，通常与 USER 相同
+#   APP_SMTP_PASSWORD — SMTP 授权码，不是邮箱登录密码！
+#     · QQ 邮箱：邮箱设置 → 账户 → POP3/IMAP/SMTP → 开启服务 → 生成授权码
+#     · 163 邮箱：设置 → POP3/SMTP/IMAP → 开启 → 设置客户端授权密码
+#     · Gmail：需开启两步验证后生成「应用专用密码」
+#
+# 【端口与加密方式】
+#   APP_SMTP_PORT=587 + APP_SMTP_USE_TLS=true  → STARTTLS（推荐，先明文握手再加密）
+#   APP_SMTP_PORT=465 + APP_SMTP_USE_TLS=false → SSL/TLS（直连加密，代码自动切换 SMTP_SSL）
+#
+#   常用邮件服务商：
+#     QQ 邮箱  smtp.qq.com      587(STARTTLS) 或 465(SSL)
+#     163 邮箱 smtp.163.com     587(STARTTLS) 或 465(SSL)
+#     Gmail    smtp.gmail.com   587(STARTTLS) 或 465(SSL)
+#     Outlook  smtp.office365.com  587(STARTTLS)
+#
+# 【频率与安全控制】
+#   APP_REGISTRATION_CODE_TTL_MINUTES    — 验证码有效期（分钟），默认 10
+#   APP_REGISTRATION_SEND_INTERVAL_SECONDS — 同一邮箱重发最短间隔（秒），默认 60
+#   APP_REGISTRATION_MAX_ATTEMPTS        — 验证码最大错误次数，超出后需重新获取，默认 5
+# ============================================================================
+APP_SMTP_HOST=smtp.qq.com
+APP_SMTP_PORT=587
+APP_SMTP_USER=your-email@qq.com
+APP_SMTP_PASSWORD=your-smtp-authorization-code
+APP_SMTP_FROM=your-email@qq.com
+APP_SMTP_USE_TLS=true
+APP_REGISTRATION_CODE_TTL_MINUTES=10
+APP_REGISTRATION_SEND_INTERVAL_SECONDS=60
+APP_REGISTRATION_MAX_ATTEMPTS=5
diff --git a/backend/error_correction_agent/__init__.py b/backend/agents/__init__.py
similarity index 100%
rename from backend/error_correction_agent/__init__.py
rename to backend/agents/__init__.py
diff --git a/backend/agents/error_correction/__init__.py b/backend/agents/error_correction/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/backend/error_correction_agent/agent.py b/backend/agents/error_correction/agent.py
similarity index 72%
rename from backend/error_correction_agent/agent.py
rename to backend/agents/error_correction/agent.py
index f1c60318..c1947199 100644
--- a/backend/error_correction_agent/agent.py
+++ b/backend/agents/error_correction/agent.py
@@ -9,8 +9,8 @@
 from langchain.agents.structured_output import ToolStrategy
 from langchain_core.messages import SystemMessage, HumanMessage
 
-from config import settings
-from llm import init_model as _init_model
+from core.config import settings
+from core.llm import init_model as _init_model
 from .prompts import SPLIT_PROMPT, SPLIT_PROMPT_LITE, CORRECTION_PROMPT
 from .schemas import QuestionSplitResult, CorrectionResult
 
@@ -21,6 +21,35 @@
 _agent_cache_lock = threading.Lock()
 
 
+def _should_avoid_tool_strategy(provider: str, model_name: str | None, cfg) -> bool:
+    """判断是否应避免 ToolStrategy（避免发送 tool_choice 到不兼容模型）"""
+    if provider != "openai":
+        return False
+
+    effective_model = (cfg.resolve_model_name(model_name) or "").lower()
+    base_url = (getattr(cfg, "base_url", "") or "").lower()
+
+    # DeepSeek 推理模型（如 deepseek-reasoner）不支持 tool_choice
+    if "reasoner" in effective_model and "deepseek" in (effective_model + base_url):
+        return True
+    return False
+
+
+def _invoke_with_json_schema_fallback(*, prompt: str, provider: str, model_name: str | None, temperature: float, schema, system_prompt: str):
+    """不使用 ToolStrategy 的结构化调用兜底路径。"""
+    model = _init_model(temperature=temperature, provider=provider, model_name=model_name)
+    messages = [
+        SystemMessage(content=system_prompt),
+        HumanMessage(content=prompt),
+    ]
+    try:
+        # 对 OpenAI-compatible（含 DeepSeek）优先使用 json_mode，避免 tool_choice。
+        return model.with_structured_output(schema, method="json_mode").invoke(messages)
+    except TypeError:
+        # 部分模型/版本不支持 method 参数。
+        return model.with_structured_output(schema).invoke(messages)
+
+
 def detect_subject_via_llm(ocr_text_sample: str, db_subjects: list, provider: str = "openai") -> str:
     """通过轻量 LLM 识别试卷科目
 
@@ -123,12 +152,18 @@ def _invoke_structured(
     """通用结构化输出调用，屏蔽 ToolStrategy vs with_structured_output 差异"""
     cfg = settings.get_provider(provider)
 
-    if not cfg.supports_function_calling:
-        model = _init_model(temperature=temperature, provider=provider, model_name=model_name)
-        return model.with_structured_output(schema).invoke([
-            SystemMessage(content=system_prompt_fallback),
-            HumanMessage(content=prompt),
-        ])
+    if (not cfg.supports_function_calling) or _should_avoid_tool_strategy(provider, model_name, cfg):
+        if _should_avoid_tool_strategy(provider, model_name, cfg):
+            effective = cfg.resolve_model_name(model_name)
+            logger.info("模型 %s 不兼容 tool_choice，切换为 JSON 结构化兜底路径", effective)
+        return _invoke_with_json_schema_fallback(
+            prompt=prompt,
+            provider=provider,
+            model_name=model_name,
+            temperature=temperature,
+            schema=schema,
+            system_prompt=system_prompt_fallback,
+        )
 
     full_key = f"{cache_key}_{provider}_{model_name or 'default'}"
     with _agent_cache_lock:
diff --git a/backend/error_correction_agent/prompts.py b/backend/agents/error_correction/prompts.py
similarity index 68%
rename from backend/error_correction_agent/prompts.py
rename to backend/agents/error_correction/prompts.py
index 5e8c5c02..c17ae3f4 100644
--- a/backend/error_correction_agent/prompts.py
+++ b/backend/agents/error_correction/prompts.py
@@ -13,9 +13,11 @@
 ## 最重要的规则（必须遵守）
 
 1. **子题禁止拆分**：含（1）（2）（3）的大题必须作为一道题输出，question_id 用大题号（如 "17"），不要输出 "17-1"、"17-2"
-2. **跨页连续性**：一道题的内容可能跨越两页。如果某页开头的内容块没有新题号（如 1. 2. 3.），它一定属于上一页最后一道题：
+2. **主处理页与上下文页**：每页带有 `is_primary` 字段。`is_primary: true` 的页必须提取其中开始的所有题目；`is_primary: false` 的页仅作跨页补全参考，不要提取只在该页开始的独立题目。
+3. **跨页连续性**：一道题的内容可能跨越两页。如果某页开头的内容块没有新题号（如 1. 2. 3.），它一定属于上一页最后一道题：
    - 页首出现 A/B/C/D 选项但无题号 → 是上一题的选项
    - 页首出现（1）（2）小问但无大题号 → 是上一题的子问内容
+   - **页首出现 image block 且其后才出现新题号 → 该图片属于上一题，加入上一题 content_blocks 末尾，不归属本页第一道题**
    - **绝对不要把这类无题号内容创建为新题目**
 3. **content_blocks 的 block_type 只能填 text 或 image**，不能填 paragraph_title 等其他值
 4. **question_type 只能填**：选择题、填空题、解答题、判断题
@@ -31,9 +33,11 @@
 ## 各字段填写
 
 - question_id：题号，如 "1"、"2"、"17"
+- section_title：该题所属的大题标题，即题目之前最近出现的 `paragraph_title` block 的文字，如"四、我会计算。"；若试卷无大题结构则填 null
 - question_type：选择题/填空题/解答题/判断题
 - content_blocks：题干内容，每项 {"block_type": "text"/"image", "content": "..."}
 - options：选择题的选项列表，其他题型填 null
+- option_images：选项对应的图片路径列表，与 options 按索引一一对应。若某个选项的内容是一张图片（如实验装置图），将图片路径填入对应位置，无图片的选项填空字符串 ""。非选择题或选项无图片时填 null
 - has_formula：含公式填 true
 - has_image：含图片填 true
 - knowledge_tags：1-3 个知识点标签，如 ["三角函数", "诱导公式"]
@@ -46,7 +50,8 @@
 
 ## 表格处理
 
-内容含表格数据时，转为 HTML 放入 text 块：`<table><thead><tr><th>...</th></tr></thead><tbody><tr><td>...</td></tr></tbody></table>`，单元格内公式仍用 LaTeX。
+- 输入 block 的 `block_label` 为 `table` 时：将其 `block_content`（已是完整 HTML）原样作为一个 `block_type: text` 的 content_block 输出，禁止拆散单元格；若 HTML 含 `<img>` 标签，路径加入 `image_refs`，设 `has_image: true`
+- 纯文本表格数据需转换时：使用 `<table><thead><tr><th>...</th></tr></thead><tbody><tr><td>...</td></tr></tbody></table>` 结构放入 text 块，单元格内公式仍用 LaTeX
 
 ## OCR 变量下标还原
 
@@ -63,89 +68,25 @@
 
 你是一个专业的试卷题目分割专家。你的任务是分析OCR识别的文档结构，智能分割出独立的题目。
 
-## 核心任务
-
-从PaddleOCR解析的结构化数据中：
-1. 识别题目边界（题号是关键标志）
-2. 提取题干内容（公式用 LaTeX 标记嵌入文本中）
-3. 识别并提取选项（如果是选择题）
-4. 识别图片块并保留引用
-5. 按阅读顺序组织内容
-
-## 题目类型识别
-
-### 1. 选择题
-- 特征：包含选项标记（A. B. C. D. 或 A) B) C) D)）
-- 需提取：题号、题干、选项列表、公式、图片
-
-### 2. 填空题
-- 特征：包含下划线 `_____` 或 `（  ）`
-- 需提取：题号、题干、公式、图片
-
-### 3. 解答题
-- 特征：要求"解答"、"证明"、"计算"等
-- 需提取：题号、题干、公式、图片
-
-### 4. 判断题
-- 特征：要求判断"正确"、"错误"或"对"、"错"
-- 需提取：题号、题干
-
-## 题号识别规则
+## 题目类型与题号
 
-常见题号格式：
-- `1.` `2.` `3.` ...
-- `1、` `2、` `3、` ...
-- `（1）` `（2）` `（3）` ...
-- `一、` `二、` `三、` ...
+**题型**（只能填以下之一）：选择题（含 A/B/C/D 选项）、填空题（含 `___` 或 `（  ）`）、解答题（计算/证明/解答）、判断题
 
-注意：
-- 题号通常在段落开头
-- 题号后跟随题干内容
-- 大题号可能包含子题号（如 1. (1) (2)）
+**题号格式**：`1.` `1、` `（1）` 等阿拉伯数字编号为小题号；`一、` `二、` 等汉字序号为大题头（见下方注意事项）
 
 ## 工作流程
 
-1. **分析block顺序**
-   - 使用 block_order 确定阅读顺序
-   - 按顺序扫描所有 block
-
-2. **识别题目起点**
-   - 查找包含题号的 text 或 paragraph_title block
-   - 记录题号和题型
-
-3. **收集题目内容**
-   - 从题号开始，收集后续的 block
-   - 遇到下一个题号时停止
-   - **content_blocks 的 block_type 只允许 `text` 或 `image` 两个值**。即使 OCR 原始数据中有 `paragraph_title`、`doc_title` 等标签，输出时一律归为 `text`
-   - 公式不单独成块，直接用 LaTeX 标记嵌入 text 的 content 中（行内 `$...$`，独占行 `$$...$$`）
-   - **图片必须嵌入 content_blocks**：所有属于该题目的 image block 都必须作为 `{"block_type": "image", "content": "/images/..."}` 加入 content_blocks，即使图片与选项关联（如实验操作图）。不要只放在 image_refs 中而不嵌入 content_blocks
-
-4. **结构化输出**
-   - 将每道题目组织为结构化数据
-   - 输出格式由系统自动约束，请按 schema 字段填写
+1. 按 block_order 顺序扫描所有 block
+2. 遇到 `paragraph_title` block（如"四、我会计算。"）时，更新当前大题标题（`section_title`）；该 block **不输出为题目**
+3. 遇到阿拉伯数字题号（1. 2. 3.）时，标记为新题目起点，并将当前 `section_title` 写入该题
+4. 收集题目内容直到下一个题号：`content_blocks` 只允许 `text`/`image`；公式用 LaTeX 嵌入 text；图片必须作为 image block 加入 content_blocks
+5. 按 schema 字段组织结构化输出
 
 ## 知识点标注
 
-为每道题目标注 1-3 个知识点标签（填写到 `knowledge_tags` 字段），要求：
-
-1. **粒度为二级知识点**：使用具体的知识点名称，不要过于宽泛也不要过于细碎
-   - 正确：`["三角函数", "诱导公式"]`
-   - 过于宽泛：`["数学"]`
-   - 过于细碎：`["三角函数的诱导公式在特殊角求值中的应用"]`
-2. **使用中文短语**，4-8 个字为宜
-3. **优先复用已有标签**：如果调用时提供了 `existing_tags`（已有知识点标签池），请优先从中选取匹配的标签，只有确实没有匹配项时才创建新标签
-4. **示例**（数学）：
-   - 复数求共轭 → `["复数", "复数运算"]`
-   - 求圆锥体积 → `["立体几何", "圆锥"]`
-   - 等差数列求和 → `["数列", "等差数列"]`
-   - 三角函数图像变换 → `["三角函数", "函数图像变换"]`
-   - 线性规划最值 → `["不等式", "线性规划"]`
-5. **示例**（物理）：
-   - 牛顿第二定律应用 → `["力学", "牛顿第二定律"]`
-   - 电场强度计算 → `["电磁学", "电场强度"]`
-6. **示例**（化学）：
-   - 离子方程式判断 → `["离子反应", "离子方程式"]`
-   - 化学平衡移动 → `["化学平衡", "平衡移动"]`
+为每道题目标注 1-3 个知识点标签（`knowledge_tags`）：使用二级知识点短语（4-8 字），有 `existing_tags` 时优先复用。
+
+示例：等差数列求和 → `["数列", "等差数列"]`；牛顿第二定律 → `["力学", "牛顿第二定律"]`；离子方程式 → `["离子反应", "离子方程式"]`
 
 ## OCR 错误标记
 
@@ -186,10 +127,20 @@
      - 含子题的大题：「17. 已知函数 f(x)…（1）求…（2）证明…」是同一道题的多问 → 输出 1 道题
    - 板块标题本身不输出为任何题目，其分值说明也无需保留
 
-4. **跨页连续性**
+4. **主处理页与上下文页（重要）**
+   - 每页数据带有 `is_primary` 字段
+   - `is_primary: true`（主处理页）：**必须**提取该页中出现的所有题目（含跨页延续到 context 页的题目）
+   - `is_primary: false`（上下文页）：**仅**用于补全主处理页题目跨页的内容，**不要**提取只在该页开始的独立题目
+   - 判断"题目在哪页开始"：题号（阿拉伯数字）首次出现在哪页，就算在哪页开始
+
+5. **跨页连续性**
    - 一道题的内容可能跨越两页。如果某页开头的内容块没有新题号，它属于上一页最后一道题
    - 典型情况：页首出现 A/B/C/D 选项但无题号 → 是上一题的选项；页首出现（1）（2）小问但无大题号 → 是上一题的子问
    - 不要把这类无题号的跨页内容创建为新题目
+   - **页首图片的归属（极其重要）**：若某页最先出现的 block 是 image，且该图片**之前没有出现任何阿拉伯数字题号**，则该图片**属于上一道题**，必须将其加入上一题的 content_blocks 末尾，而不是归入本页第一道题。
+     - 判断依据：图片归属看的是"该图片出现在哪道题的题号之后"。图片在题号之前 → 归上一题；图片在题号之后 → 归该题。
+     - 即使本页第一道题的题干中含"如图所示"，也不能把题号之前出现的图片归给它。
+     - 示例：页面顺序为 `[image: 电路图] → [text: "21. 桔槔是..."]`，该电路图在题号 21 之前出现，应归属题号 20（上一题），而非题目 21。
 
 5. **准确性优先**
    - 不确定的边界，宁可保守（保留更多内容）
@@ -217,14 +168,19 @@
    - 对于选择题，如果你已经把选项提取到 `options` 数组中，则必须从 `content_blocks` 的文本中移除选项部分
    - `content_blocks` 只保留题干（题目正文），不包含选项文本
    - 例如原始文本为 "下列说法正确的是（）\\nA. xxx B. yyy"，则 content_blocks 中的 text 只保留 "下列说法正确的是（）"，选项 "A. xxx"、"B. yyy" 放入 options 数组
+   - **所有选项必须完整提取**：选择题通常有 4 个选项（A/B/C/D），务必逐一检查，不要遗漏任何选项。如果某个选项的 OCR 文本与图片分散在不同 block 中，仍需完整识别并提取
 
 9. **表格处理（重要）**
-   - 如果题目内容中包含表格数据（如实验数据、统计数据、对比数据），必须将其转换为 HTML 表格
-   - 使用如下结构，放入 `text` 类型的 content_block：
-     `<table><thead><tr><th>列1</th><th>列2</th></tr></thead><tbody><tr><td>值1</td><td>值2</td></tr></tbody></table>`
-   - 表格单元格内的公式仍然用 LaTeX 标记（如 `<td>$x_i$</td>`）
-   - 每行数据对应一个 `<tr>`，表头行放在 `<thead>` 中，数据行放在 `<tbody>` 中
-   - 例如：试验序号/伸缩率数据应输出为包含 `<table>` 的 HTML 字符串，而不是纯文本
+
+   **情况一：输入 block 已是 table 类型（`block_label: "table"`）**
+   - 其 `block_content` 已是完整 HTML，**必须将整段 HTML 原样作为一个 `block_type: text` 的 content_block 输出**
+   - 禁止拆散单元格，禁止把表格中的图片单独提取为独立的 image block
+   - 若 HTML 中含有 `<img>` 标签，将所有 src 路径加入 `image_refs`，并设 `has_image: true`
+
+   **情况二：纯文本表格数据需要转换**
+   - 如果题目含有表格数据但输入为纯文本（如实验数据、统计数据、对比数据），将其转换为 HTML 表格，放入 `text` 类型的 content_block
+   - 使用结构：`<table><thead><tr><th>列1</th><th>列2</th></tr></thead><tbody><tr><td>值1</td><td>值2</td></tr></tbody></table>`
+   - 表格单元格内的公式仍用 LaTeX 标记（如 `<td>$x_i$</td>`）
 
 ## 示例
 
diff --git a/backend/error_correction_agent/schemas.py b/backend/agents/error_correction/schemas.py
similarity index 80%
rename from backend/error_correction_agent/schemas.py
rename to backend/agents/error_correction/schemas.py
index 5bff6bb6..ae7dc6b6 100644
--- a/backend/error_correction_agent/schemas.py
+++ b/backend/agents/error_correction/schemas.py
@@ -16,9 +16,11 @@ class ContentBlock(BaseModel):
 class Question(BaseModel):
     """单道题目"""
     question_id: str = Field(description="题号，如 '1', '2', '(1)' 等")
+    section_title: Optional[str] = Field(default=None, description="该题所属的大题标题，即题目之前最近出现的 paragraph_title block 的文字，如'四、我会计算。'；若试卷无大题结构则为 null")
     question_type: Literal["选择题", "填空题", "解答题", "判断题"] = Field(description="题目类型")
     content_blocks: List[ContentBlock] = Field(description="题干内容块列表（不含选项）")
     options: Optional[List[str]] = Field(default=None, description="选项列表，仅选择题需要，如 ['A. xxx', 'B. yyy']")
+    option_images: Optional[List[str]] = Field(default=None, description="选项对应的图片路径列表，与 options 按索引一一对应；无图片的选项填 null 或空字符串")
     has_formula: bool = Field(default=False, description="是否包含数学公式")
     has_image: bool = Field(default=False, description="是否包含图片")
     image_refs: Optional[List[str]] = Field(default=None, description="图片引用路径列表")
@@ -35,9 +37,11 @@ class QuestionSplitResult(BaseModel):
 class CorrectedQuestion(BaseModel):
     """纠错后的单道题目"""
     question_id: str = Field(description="题号")
+    section_title: Optional[str] = Field(default=None, description="所属大题标题，透传自分割结果，不需要修改")
     question_type: Literal["选择题", "填空题", "解答题", "判断题"] = Field(description="题目类型")
     content_blocks: List[ContentBlock] = Field(description="纠错后的题干内容块列表")
     options: Optional[List[str]] = Field(default=None, description="纠错后的选项列表")
+    option_images: Optional[List[str]] = Field(default=None, description="选项对应的图片路径列表，与 options 按索引一一对应")
     has_formula: bool = Field(default=False, description="是否包含数学公式")
     has_image: bool = Field(default=False, description="是否包含图片")
     image_refs: Optional[List[str]] = Field(default=None, description="图片引用路径列表")
diff --git a/backend/error_correction_agent/tools/__init__.py b/backend/agents/error_correction/tools/__init__.py
similarity index 100%
rename from backend/error_correction_agent/tools/__init__.py
rename to backend/agents/error_correction/tools/__init__.py
diff --git a/backend/error_correction_agent/tools/file_tools.py b/backend/agents/error_correction/tools/file_tools.py
similarity index 100%
rename from backend/error_correction_agent/tools/file_tools.py
rename to backend/agents/error_correction/tools/file_tools.py
diff --git a/backend/error_correction_agent/tools/question_tools.py b/backend/agents/error_correction/tools/question_tools.py
similarity index 99%
rename from backend/error_correction_agent/tools/question_tools.py
rename to backend/agents/error_correction/tools/question_tools.py
index aa1943d1..37b95fd4 100644
--- a/backend/error_correction_agent/tools/question_tools.py
+++ b/backend/agents/error_correction/tools/question_tools.py
@@ -9,7 +9,7 @@
 from typing import List, Dict, Any
 from dotenv import load_dotenv
 from langchain_core.tools import tool
-from config import settings
+from core.config import settings
 from src.utils import simplify_ocr_results, run_async
 
 load_dotenv()
diff --git a/backend/agents/note/__init__.py b/backend/agents/note/__init__.py
new file mode 100644
index 00000000..cc59c13f
--- /dev/null
+++ b/backend/agents/note/__init__.py
@@ -0,0 +1 @@
+from .agent import invoke_note_organize
diff --git a/backend/agents/note/agent.py b/backend/agents/note/agent.py
new file mode 100644
index 00000000..7ef827da
--- /dev/null
+++ b/backend/agents/note/agent.py
@@ -0,0 +1,136 @@
+"""
+笔记整理 Agent — 将 OCR 文本整理为结构化 Markdown 笔记
+
+支持：
+  - 重试机制（指数退避，最多 3 次）
+  - 多页分批处理（超过阈值时分批整理再合并）
+"""
+
+import logging
+import time
+
+from langchain_core.messages import SystemMessage, HumanMessage
+
+from core.config import settings
+from core.llm import init_model
+from .prompts import NOTE_ORGANIZE_PROMPT
+from .schemas import OrganizedNote
+
+logger = logging.getLogger(__name__)
+
+# 单次 LLM 调用的文本长度上限（字符数），超过则分批
+BATCH_CHAR_LIMIT = 6000
+MAX_RETRIES = 3
+
+
+def _invoke_once(model, ocr_text: str) -> OrganizedNote:
+    """单次 LLM 调用，返回结构化笔记"""
+    structured_model = model.with_structured_output(OrganizedNote, method="function_calling")
+    return structured_model.invoke([
+        SystemMessage(content=NOTE_ORGANIZE_PROMPT),
+        HumanMessage(content=f"以下是 OCR 识别出的课堂笔记原始文本，请整理为结构化笔记：\n\n{ocr_text}"),
+    ])
+
+
+def _invoke_with_retry(model, ocr_text: str) -> OrganizedNote:
+    """带指数退避的重试调用"""
+    last_error = None
+    for attempt in range(1, MAX_RETRIES + 1):
+        try:
+            result = _invoke_once(model, ocr_text)
+            if result:
+                return result
+            last_error = "LLM 未返回有效结果"
+        except Exception as e:
+            last_error = str(e)
+            logger.warning(f"笔记整理: 第 {attempt}/{MAX_RETRIES} 次失败: {last_error}")
+
+        if attempt < MAX_RETRIES:
+            wait = 2 ** attempt  # 2s, 4s
+            logger.info(f"笔记整理: {wait}s 后重试...")
+            time.sleep(wait)
+
+    raise RuntimeError(f"笔记整理失败（重试 {MAX_RETRIES} 次）: {last_error}")
+
+
+def _split_text_into_batches(text: str) -> list:
+    """按行将文本拆分为多个批次，每批不超过 BATCH_CHAR_LIMIT 字符"""
+    lines = text.split('\n')
+    batches = []
+    current = []
+    current_len = 0
+
+    for line in lines:
+        if current_len + len(line) + 1 > BATCH_CHAR_LIMIT and current:
+            batches.append('\n'.join(current))
+            current = []
+            current_len = 0
+        current.append(line)
+        current_len += len(line) + 1
+
+    if current:
+        batches.append('\n'.join(current))
+    return batches
+
+
+def _merge_notes(notes: list) -> OrganizedNote:
+    """合并多批次的笔记结果"""
+    # 用第一批的标题和科目
+    title = notes[0].title
+    subject = notes[0].subject
+
+    # 合并 Markdown 内容
+    all_markdown = '\n\n'.join(n.content_markdown for n in notes)
+
+    # 合并去重标签
+    all_tags = []
+    seen = set()
+    for n in notes:
+        for tag in n.knowledge_tags:
+            if tag not in seen:
+                all_tags.append(tag)
+                seen.add(tag)
+
+    return OrganizedNote(
+        title=title,
+        subject=subject,
+        content_markdown=all_markdown,
+        knowledge_tags=all_tags[:5],  # 最多 5 个
+    )
+
+
+def invoke_note_organize(
+    ocr_text: str,
+    provider: str = "openai",
+    model_name: str = None,
+) -> OrganizedNote:
+    """调用 LLM 将 OCR 文本整理为结构化笔记
+
+    文本较短时直接整理，较长时分批处理再合并。
+    每次调用带 3 次重试（指数退避）。
+
+    Args:
+        ocr_text: OCR 识别出的原始文本
+        provider: 模型供应商
+        model_name: 指定模型名称，None 使用默认
+
+    Returns:
+        OrganizedNote 结构化结果
+    """
+    model = init_model(temperature=0.3, provider=provider, model_name=model_name)
+
+    # 短文本：直接整理
+    if len(ocr_text) <= BATCH_CHAR_LIMIT:
+        return _invoke_with_retry(model, ocr_text)
+
+    # 长文本：分批处理
+    batches = _split_text_into_batches(ocr_text)
+    logger.info(f"笔记文本较长（{len(ocr_text)} 字符），分 {len(batches)} 批处理")
+
+    results = []
+    for i, batch in enumerate(batches):
+        logger.info(f"处理第 {i + 1}/{len(batches)} 批（{len(batch)} 字符）")
+        result = _invoke_with_retry(model, batch)
+        results.append(result)
+
+    return _merge_notes(results)
diff --git a/backend/agents/note/prompts.py b/backend/agents/note/prompts.py
new file mode 100644
index 00000000..09a9d270
--- /dev/null
+++ b/backend/agents/note/prompts.py
@@ -0,0 +1,52 @@
+"""
+笔记整理 Prompt
+"""
+
+NOTE_ORGANIZE_PROMPT = """你是一位专业的学习笔记整理助手。
+
+## 任务
+将 OCR 识别出的课堂笔记原始文本，整理为结构清晰、条理分明的 Markdown 格式知识点笔记。
+
+## 输入
+用户会提供 OCR 识别出的原始文本，可能包含：
+- 手写笔记内容
+- 板书内容
+- 课堂重点标注
+- 公式、定理、定义
+- 例题和解题步骤
+
+## 整理要求
+
+### 标题
+- 根据笔记内容自动生成一个简洁的标题
+- 格式：【科目】主题，如 "高中数学 - 三角函数基本性质"
+
+### 内容整理
+1. **识别知识结构**：提取定义、定理、公式、性质、方法等
+2. **层级组织**：用 Markdown 标题（##、###）组织层级关系
+3. **公式还原**：将 OCR 可能识别不完整的数学公式用 LaTeX 标记还原（行内 $...$，独占行 $$...$$）
+4. **要点标注**：用 **加粗** 标注重点，用 > 引用块标注注意事项
+5. **列表归纳**：零散的知识点用有序/无序列表归纳
+6. **纠正错误**：修正 OCR 识别中的明显错误（如乱码、错别字、符号错误）
+
+### 知识点标签
+- 提取 2-5 个核心知识点作为标签
+- 标签用中文，4-8 个字，如 "三角函数"、"诱导公式"
+- 优先使用通用教材中的标准术语
+
+### 科目识别
+- 根据内容判断科目，格式：学段+科目（如 "高中数学"、"初中物理"）
+- 无法判断时输出 "未知"
+
+### 图片处理
+- 输入文本中的 `[图片: /images/xxx.jpg]` 表示 OCR 识别到的图片（如几何图形、示意图、图表等）
+- 在 Markdown 中对应位置插入图片引用：`![图片描述](/images/xxx.jpg)`
+- 根据上下文给图片添加合适的描述文字（如 "平行四边形示意图"、"函数图像" 等）
+- 不要遗漏任何图片
+
+## 注意
+- 保留原始笔记中有价值的内容，不要遗漏
+- 不要添加原文中没有的知识内容
+- 如果原文有例题，保留例题和解法
+- 输出纯 Markdown，不要包含 ```markdown``` 代码块标记
+"""
diff --git a/backend/agents/note/schemas.py b/backend/agents/note/schemas.py
new file mode 100644
index 00000000..ef036f05
--- /dev/null
+++ b/backend/agents/note/schemas.py
@@ -0,0 +1,14 @@
+"""
+笔记整理结构化输出 Schema
+"""
+
+from typing import List, Optional
+from pydantic import BaseModel, Field
+
+
+class OrganizedNote(BaseModel):
+    """LLM 整理后的笔记结构"""
+    title: str = Field(description="笔记标题，根据内容自动生成，简洁明确")
+    subject: str = Field(description="科目，如 '高中数学'、'初中物理'，无法判断则输出 '未知'")
+    content_markdown: str = Field(description="整理后的 Markdown 格式笔记内容，包含知识点归纳、公式、要点等")
+    knowledge_tags: List[str] = Field(description="知识点标签列表（2-5 个），如 ['三角函数', '诱导公式']")
diff --git a/backend/solve_agent/__init__.py b/backend/agents/solve/__init__.py
similarity index 100%
rename from backend/solve_agent/__init__.py
rename to backend/agents/solve/__init__.py
diff --git a/backend/solve_agent/agent.py b/backend/agents/solve/agent.py
similarity index 98%
rename from backend/solve_agent/agent.py
rename to backend/agents/solve/agent.py
index bfa233d0..478b0fcb 100644
--- a/backend/solve_agent/agent.py
+++ b/backend/agents/solve/agent.py
@@ -11,7 +11,7 @@
 
 from langchain_core.messages import SystemMessage, HumanMessage
 
-from llm import init_model
+from core.llm import init_model
 from .prompts import SOLVE_PROMPT
 from .schemas import SolveBatchResult
 
diff --git a/backend/solve_agent/prompts.py b/backend/agents/solve/prompts.py
similarity index 100%
rename from backend/solve_agent/prompts.py
rename to backend/agents/solve/prompts.py
diff --git a/backend/solve_agent/schemas.py b/backend/agents/solve/schemas.py
similarity index 100%
rename from backend/solve_agent/schemas.py
rename to backend/agents/solve/schemas.py
diff --git a/backend/teach_agent/__init__.py b/backend/agents/teach/__init__.py
similarity index 100%
rename from backend/teach_agent/__init__.py
rename to backend/agents/teach/__init__.py
diff --git a/backend/teach_agent/agent.py b/backend/agents/teach/agent.py
similarity index 55%
rename from backend/teach_agent/agent.py
rename to backend/agents/teach/agent.py
index e8765051..67aca635 100644
--- a/backend/teach_agent/agent.py
+++ b/backend/agents/teach/agent.py
@@ -7,7 +7,7 @@
 
 from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
 
-from llm import init_model
+from core.llm import init_model
 from .prompts import build_teach_prompt
 
 logger = logging.getLogger(__name__)
@@ -37,9 +37,15 @@ def _build_question_text(question: Dict[str, Any]) -> str:
     return "\n".join(parts)
 
 
+GENERAL_SYSTEM_PROMPT = """你是一位专业的学习辅导助手，擅长数学、物理、化学、英语等学科。
+请用清晰、有条理的方式回答学生的问题。
+如果涉及数学公式，使用 LaTeX 标记（行内 $...$，独占行 $$...$$）。
+回答要准确、简洁，适合中学生和大学生理解。"""
+
+
 def stream_teach(
     *,
-    question: Dict[str, Any],
+    question: Dict[str, Any] = None,
     messages: List[Dict[str, str]],
     provider: str = "openai",
     model_name: str | None = None,
@@ -47,8 +53,7 @@ def stream_teach(
     """流式教学对话
 
     Args:
-        question: 题目数据（含 content_json, options_json, answer, question_type,
-                  subject, knowledge_tags 等字段）
+        question: 题目数据（可选，None 为独立对话）
         messages: 对话历史 [{"role": "user"|"assistant", "content": "..."}]
         provider: 模型供应商
         model_name: 指定模型名称，为 None 时使用 provider 默认模型
@@ -56,17 +61,20 @@ def stream_teach(
     Yields:
         逐 token 的文本片段
     """
-    subject = question.get("subject") or "未知科目"
-    knowledge_tags = question.get("knowledge_tags") or []
-    answer_text = question.get("answer") or "暂无答案"
-    question_text = _build_question_text(question)
-
-    system_prompt = build_teach_prompt(
-        subject=subject,
-        knowledge_tags=knowledge_tags,
-        question_text=question_text,
-        answer_text=answer_text,
-    )
+    if question:
+        subject = question.get("subject") or "未知科目"
+        knowledge_tags = question.get("knowledge_tags") or []
+        answer_text = question.get("answer") or "暂无答案"
+        question_text = _build_question_text(question)
+
+        system_prompt = build_teach_prompt(
+            subject=subject,
+            knowledge_tags=knowledge_tags,
+            question_text=question_text,
+            answer_text=answer_text,
+        )
+    else:
+        system_prompt = GENERAL_SYSTEM_PROMPT
 
     # 构建 LangChain 消息列表
     lc_messages = [SystemMessage(content=system_prompt)]
@@ -79,6 +87,16 @@ def stream_teach(
     model = init_model(temperature=0.4, provider=provider, model_name=model_name)
 
     for chunk in model.stream(lc_messages):
+        # DeepSeek reasoner 模型会在 additional_kwargs 里返回 reasoning_content
+        reasoning = None
+        if hasattr(chunk, 'additional_kwargs'):
+            reasoning = chunk.additional_kwargs.get('reasoning_content')
+
         token = chunk.content
-        if token:
-            yield token
+        # 安全处理：确保 token 是字符串（某些模型返回 list）
+        if isinstance(token, list):
+            token = ''.join(str(t) for t in token if t)
+        if reasoning and isinstance(reasoning, str):
+            yield {"type": "reasoning", "content": reasoning}
+        if token and isinstance(token, str):
+            yield {"type": "content", "content": token}
diff --git a/backend/teach_agent/prompts.py b/backend/agents/teach/prompts.py
similarity index 100%
rename from backend/teach_agent/prompts.py
rename to backend/agents/teach/prompts.py
diff --git a/backend/benchmark/collect.py b/backend/benchmark/collect.py
index 7026d402..68d2334a 100644
--- a/backend/benchmark/collect.py
+++ b/backend/benchmark/collect.py
@@ -29,7 +29,7 @@
 if str(BACKEND_ROOT) not in sys.path:
     sys.path.insert(0, str(BACKEND_ROOT))
 
-from config import settings
+from core.config import settings
 
 logger = logging.getLogger(__name__)
 
@@ -93,7 +93,7 @@ def ocr_and_split(image_paths: List[str], provider: str = "openai") -> List[Dict
     batches = _build_overlapping_batches(ocr_data, batch_size=2, overlap=1)
 
     # 分割
-    from error_correction_agent.tools import split_batch
+    from agents.error_correction.tools import split_batch
 
     all_questions = []
     for batch_data in batches:
diff --git a/backend/benchmark/evaluate.py b/backend/benchmark/evaluate.py
index b9c64023..eca4e4ae 100644
--- a/backend/benchmark/evaluate.py
+++ b/backend/benchmark/evaluate.py
@@ -62,7 +62,7 @@ def run_evaluation(provider: str, targets: List[Dict[str, Any]]) -> Dict[str, An
     Returns:
         评测报告 dict
     """
-    from solve_agent import invoke_solve
+    from agents.solve import invoke_solve
 
     # 提取题目数据
     questions = []
diff --git a/backend/core/__init__.py b/backend/core/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/backend/config.py b/backend/core/config.py
similarity index 94%
rename from backend/config.py
rename to backend/core/config.py
index dd11c106..f6c37683 100644
--- a/backend/config.py
+++ b/backend/core/config.py
@@ -12,7 +12,7 @@
 _CONNECTION_CACHE_TTL = 60  # 连通性检测缓存有效期（秒）
 _providers_lock = threading.Lock()  # 保护 llm_providers 并发读写
 
-_BACKEND_ROOT = Path(__file__).resolve().parent
+_BACKEND_ROOT = Path(__file__).resolve().parent.parent  # backend/core/ → backend/
 _PROJECT_ROOT = _BACKEND_ROOT.parent
 _ENV_FILE = _PROJECT_ROOT / ".env"
 
@@ -180,6 +180,17 @@ class Settings(BaseSettings):
     max_file_size_mb: int = 50
     allowed_extensions: set[str] = {"pdf", "png", "jpg", "jpeg", "bmp", "tiff", "webp"}
 
+    # 注册验证码邮件（SMTP，环境变量前缀仍为 APP_，如 APP_SMTP_HOST）
+    smtp_host: str = ""
+    smtp_port: int = 587
+    smtp_user: str = ""
+    smtp_password: str = ""
+    smtp_from: str = ""
+    smtp_use_tls: bool = True
+    registration_code_ttl_minutes: int = 10
+    registration_send_interval_seconds: int = 60
+    registration_max_attempts: int = 5
+
     # LLM 供应商注册表
     llm_providers: dict[str, LLMProviderConfig] | None = None
 
@@ -198,7 +209,7 @@ def _resolve_defaults(self):
         if self.erased_dir is None:
             self.erased_dir = self.runtime_dir / "erased"
         if self.model_path is None:
-            self.model_path = self.runtime_dir / "models" / "latest.pth"
+            self.model_path = _BACKEND_ROOT / "models" / "weight" / "best.pth"
 
         # 初始化 LLM 供应商注册表
         if self.llm_providers is None:
@@ -271,7 +282,7 @@ def load_providers_from_db(self, user_id: int) -> None:
     def _clear_agent_cache(self):
         """清除 agent 缓存"""
         try:
-            from error_correction_agent.agent import _agent_cache, _agent_cache_lock
+            from agents.error_correction.agent import _agent_cache, _agent_cache_lock
             with _agent_cache_lock:
                 _agent_cache.clear()
         except ImportError:
diff --git a/backend/llm.py b/backend/core/llm.py
similarity index 94%
rename from backend/llm.py
rename to backend/core/llm.py
index e361ed6a..91c44865 100644
--- a/backend/llm.py
+++ b/backend/core/llm.py
@@ -5,7 +5,7 @@
 供应商配置和工厂方法统一在 config.py 的 LLMProviderConfig 子类中定义。
 """
 
-from config import settings
+from core.config import settings
 
 
 def init_model(
@@ -33,7 +33,7 @@ def init_model(
     """
     if api_key:
         # 用传入的凭据动态构建配置，跳过环境变量
-        from config import OpenAICompatibleConfig, AnthropicCompatibleConfig
+        from core.config import OpenAICompatibleConfig, AnthropicCompatibleConfig
         if provider == "anthropic":
             cfg = AnthropicCompatibleConfig(
                 api_key=api_key,
diff --git a/backend/core/mail.py b/backend/core/mail.py
new file mode 100644
index 00000000..8e21fbe1
--- /dev/null
+++ b/backend/core/mail.py
@@ -0,0 +1,59 @@
+"""SMTP 发信（注册验证码等）。"""
+
+import logging
+import smtplib
+import threading
+from email.mime.text import MIMEText
+
+logger = logging.getLogger(__name__)
+
+
+def _send_sync(
+    host, port, user, password, mail_from, use_tls, to_addr, msg,
+):
+    if port == 465:
+        smtp = smtplib.SMTP_SSL(host, port, timeout=15)
+    else:
+        smtp = smtplib.SMTP(host, port, timeout=15)
+        if use_tls:
+            smtp.starttls()
+
+    with smtp:
+        if user and password:
+            smtp.login(user, password)
+        smtp.sendmail(mail_from, [to_addr], msg.as_string())
+
+    logger.info("已发送邮件至 %s", to_addr)
+
+
+def send_smtp_email(
+    *,
+    host: str,
+    port: int,
+    user: str,
+    password: str,
+    mail_from: str,
+    use_tls: bool,
+    to_addr: str,
+    subject: str,
+    body: str,
+    async_send: bool = True,
+) -> None:
+    if not host or not mail_from:
+        raise ValueError("SMTP 未配置：需要 smtp_host 与 smtp_from")
+
+    msg = MIMEText(body, "plain", "utf-8")
+    msg["Subject"] = subject
+    msg["From"] = mail_from
+    msg["To"] = to_addr
+
+    if async_send:
+        t = threading.Thread(
+            target=_send_sync,
+            args=(host, port, user, password, mail_from, use_tls, to_addr, msg),
+            daemon=True,
+        )
+        t.start()
+        logger.info("邮件发送任务已提交（异步）: %s", to_addr)
+    else:
+        _send_sync(host, port, user, password, mail_from, use_tls, to_addr, msg)
diff --git a/backend/core/state.py b/backend/core/state.py
new file mode 100644
index 00000000..1e3d85ab
--- /dev/null
+++ b/backend/core/state.py
@@ -0,0 +1,42 @@
+"""
+全局会话状态 — 供各 Blueprint 模块共享
+
+所有需要跨路由访问的运行时变量集中在此模块，
+通过 session_lock 保护并发写操作。
+
+会话状态按 user_id 隔离，每个用户拥有独立的上传/分割上下文。
+"""
+
+import threading
+
+from src.workflow import build_workflow
+
+# 全局工作流图（无状态，可安全共享）
+workflow_graph = build_workflow()
+
+# 线程锁（全局共享）
+session_lock = threading.Lock()
+
+# ── 按用户隔离的会话状态 ──────────────────────────────────
+
+_user_sessions: dict = {}
+
+
+def get_user_session(user_id) -> dict:
+    """获取或创建指定用户的会话状态"""
+    key = user_id if user_id is not None else "anon"
+    if key not in _user_sessions:
+        _user_sessions[key] = {
+            "current_thread_id": None,
+            "session_files": {},          # file_key → {filename, filepath}
+            "session_file_order": [],     # 上传顺序
+            "cancelled_file_keys": set(), # 已取消的文件 key
+            "erased_file_paths": None,    # 擦除后的文件路径列表
+        }
+    return _user_sessions[key]
+
+
+def clear_user_session(user_id):
+    """清除指定用户的会话状态"""
+    key = user_id if user_id is not None else "anon"
+    _user_sessions.pop(key, None)
diff --git a/backend/db/__init__.py b/backend/db/__init__.py
index e53e5ae7..1904304c 100644
--- a/backend/db/__init__.py
+++ b/backend/db/__init__.py
@@ -9,7 +9,7 @@
 
 # 添加 backend 目录到路径以支持导入 config
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-from config import settings
+from core.config import settings
 
 # 确保数据库目录存在
 db_dir = settings.db_path.parent
@@ -41,6 +41,16 @@ def _migrate_schema():
         if 'answer' not in columns:
             cursor.execute("ALTER TABLE questions ADD COLUMN answer TEXT")
             conn.commit()
+
+        # chat_sessions 表：question_id 需要改为 nullable
+        # SQLite 不支持 ALTER COLUMN，需要重建表
+        cursor.execute("PRAGMA table_info(chat_sessions)")
+        cs_columns = {row[1]: row for row in cursor.fetchall()}
+        if 'title' not in cs_columns:
+            # 旧表结构，需要重建
+            cursor.execute("DROP TABLE IF EXISTS chat_messages")
+            cursor.execute("DROP TABLE IF EXISTS chat_sessions")
+            conn.commit()
     finally:
         conn.close()
 
diff --git a/backend/db/crud.py b/backend/db/crud.py
deleted file mode 100644
index e79a56ed..00000000
--- a/backend/db/crud.py
+++ /dev/null
@@ -1,1102 +0,0 @@
-"""
-数据库 CRUD 操作
-"""
-
-import hashlib
-import json
-import logging
-import re
-from datetime import datetime
-from typing import List, Dict, Any, Optional, Tuple
-
-from sqlalchemy.orm import Session, joinedload, selectinload
-from sqlalchemy import func
-
-logger = logging.getLogger(__name__)
-
-from db.models import UploadBatch, Question, KnowledgeTag, QuestionTagMapping, ChatSession, ChatMessage, SplitRecord, User, ProviderConfig
-
-
-def _filter_by_subject(query, subject: Optional[str]):
-    """如有学科筛选，为 Question 查询追加 JOIN UploadBatch 过滤"""
-    if subject:
-        return query.join(UploadBatch).filter(UploadBatch.subject == subject)
-    return query
-
-
-
-def _filter_by_user(query, user_id):
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-    return query
-
-
-# ============================================================
-# 用户认证 CRUD
-# ============================================================
-
-def create_user(db, email, password_hash, username, is_admin=False):
-    """创建用户"""
-    user = User(email=email, password_hash=password_hash, username=username, is_admin=is_admin)
-    db.add(user)
-    try:
-        db.commit()
-        db.refresh(user)
-        return user
-    except Exception as e:
-        db.rollback()
-        raise
-
-
-def get_user_by_email(db, email):
-    """按邮箱查询用户"""
-    return db.query(User).filter(User.email == email).first()
-
-
-def get_user_by_id(db, user_id):
-    """按 ID 查询用户"""
-    return db.query(User).filter(User.id == user_id).first()
-
-
-def get_user_by_login(db, identifier):
-    """按邮箱或用户名查询用户（登录用，大小写不敏感邮箱）"""
-    identifier = identifier.strip()
-    return db.query(User).filter(
-        (func.lower(User.email) == identifier.lower()) | (User.username == identifier)
-    ).first()
-
-
-def _parse_tag_list(knowledge_tag: str) -> List[str]:
-    """将逗号分隔的标签字符串拆分为去空白的非空列表"""
-    return [t.strip() for t in knowledge_tag.split(',') if t.strip()]
-
-
-def compute_content_hash(content_blocks: List[Dict]) -> str:
-    """
-    基于 content_blocks 计算去重哈希
-    使用题目文本内容计算 SHA256
-    """
-    text_parts = []
-    for block in content_blocks:
-        if block.get("block_type") == "text":
-            text_parts.append(block.get("content", ""))
-
-    text = " ".join(text_parts).strip()
-    if not text:
-        # 如果没有文本内容，使用整个 content_blocks 的 JSON 作为哈希源
-        text = json.dumps(content_blocks, ensure_ascii=False)
-
-    return hashlib.sha256(text.encode()).hexdigest()
-
-
-def get_existing_subjects(db, user_id=None):
-    """获取数据库中已有的所有科目名称（去重）"""
-    query = db.query(UploadBatch.subject).distinct().filter(
-        UploadBatch.subject.isnot(None),
-        UploadBatch.subject != "",
-    )
-    if user_id is not None:
-        query = query.filter(UploadBatch.user_id == user_id)
-    return [r[0] for r in query.all()]
-
-
-def get_existing_tag_names(db: Session, subject: Optional[str] = None) -> List[str]:
-    """获取数据库中已有的知识点标签名称列表（字符串）"""
-    query = db.query(KnowledgeTag.tag_name).distinct()
-    if subject:
-        query = query.filter(KnowledgeTag.subject == subject)
-    rows = query.order_by(KnowledgeTag.tag_name).all()
-    return [r[0] for r in rows]
-
-
-def get_or_create_tag(db: Session, tag_name: str, subject: str) -> KnowledgeTag:
-    """获取或创建知识点标签"""
-    tag = db.query(KnowledgeTag).filter_by(
-        tag_name=tag_name,
-        subject=subject
-    ).first()
-
-    if not tag:
-        tag = KnowledgeTag(tag_name=tag_name, subject=subject)
-        db.add(tag)
-        db.flush()
-
-    return tag
-
-
-def question_exists(db, content_hash, user_id=None):
-    """检查题目是否已存在（通过内容哈希 + 用户隔离）"""
-    q = db.query(Question).filter(Question.content_hash == content_hash)
-    if user_id is not None:
-        q = q.filter(Question.user_id == user_id)
-    return q.first()
-
-
-def save_questions_to_db(
-    db: Session,
-    questions: List[Dict],
-    batch_info: Dict[str, Any],
-    user_id=None,
-) -> Dict[str, int]:
-    """
-    批量入库题目
-
-    Args:
-        db: 数据库会话
-        questions: 题目列表（字典格式，来自 questions.json）
-        batch_info: 批次信息，包含 original_filename, file_path 等
-
-    Returns:
-        dict: {"created": 新增数量, "duplicates": 重复数量}
-    """
-    # 科目由编排智能体识别，不再使用关键词匹配
-    subject = batch_info.get("subject") or "未知"
-
-    # 创建批次记录
-    batch = UploadBatch(
-        user_id=user_id,
-        original_filename=batch_info.get("original_filename", "未知"),
-        subject=subject,
-        file_path=batch_info.get("file_path", ""),
-        upload_time=batch_info.get("upload_time") or datetime.utcnow()
-    )
-    db.add(batch)
-    db.flush()  # 获取 batch.id
-
-    created = 0
-    duplicates = 0
-
-    for q in questions:
-        content_blocks = q.get("content_blocks", [])
-        if not content_blocks:
-            continue
-
-        content_hash = compute_content_hash(content_blocks)
-
-        # 检查是否已存在
-        if question_exists(db, content_hash, user_id=user_id):
-            duplicates += 1
-            continue
-
-        # 创建题目记录
-        question = Question(
-            user_id=user_id,
-            batch_id=batch.id,
-            content_hash=content_hash,
-            question_type=q.get("question_type"),
-            content_json=json.dumps(content_blocks, ensure_ascii=False),
-            options_json=json.dumps(q.get("options"), ensure_ascii=False) if q.get("options") else None,
-            has_formula=q.get("has_formula", False),
-            has_image=q.get("has_image", False),
-            image_refs_json=json.dumps(q.get("image_refs"), ensure_ascii=False) if q.get("image_refs") else None,
-            needs_correction=q.get("needs_correction", False),
-            ocr_issues_json=json.dumps(q.get("ocr_issues"), ensure_ascii=False) if q.get("ocr_issues") else None,
-            answer=q.get("answer") or None,
-            user_answer=q.get("user_answer") or None,
-        )
-        db.add(question)
-        db.flush()
-
-        # 处理知识点标签
-        knowledge_tags = q.get("knowledge_tags") or []
-        for tag_name in knowledge_tags:
-            tag = get_or_create_tag(db, tag_name, subject)
-            mapping = QuestionTagMapping(
-                question_id=question.id,
-                tag_id=tag.id
-            )
-            db.add(mapping)
-
-        created += 1
-
-    db.commit()
-
-    return {"created": created, "duplicates": duplicates}
-
-
-def get_questions_by_subject(
-    db: Session,
-    subject: str,
-    limit: int = 100,
-    offset: int = 0
-) -> List[Question]:
-    """按科目查询题目"""
-    return db.query(Question).join(UploadBatch).filter(
-        UploadBatch.subject == subject
-    ).order_by(Question.created_at.desc()).offset(offset).limit(limit).all()
-
-
-def get_questions_by_tag(
-    db: Session,
-    tag_name: str,
-    limit: int = 100,
-    offset: int = 0
-) -> List[Question]:
-    """按标签查询题目"""
-    return db.query(Question).join(QuestionTagMapping).join(KnowledgeTag).filter(
-        KnowledgeTag.tag_name == tag_name
-    ).order_by(Question.created_at.desc()).offset(offset).limit(limit).all()
-
-
-def get_all_tags(db: Session, subject: Optional[str] = None) -> List[KnowledgeTag]:
-    """获取所有标签（可按科目筛选）"""
-    query = db.query(KnowledgeTag)
-    if subject:
-        query = query.filter(KnowledgeTag.subject == subject)
-    return query.order_by(KnowledgeTag.tag_name).all()
-
-
-def get_statistics(db: Session, subject: Optional[str] = None, user_id=None) -> Dict[str, Any]:
-    """获取统计信息"""
-    q_query = _filter_by_subject(db.query(func.count(Question.id)), subject)
-    if user_id is not None:
-        q_query = q_query.filter(Question.user_id == user_id)
-    total_questions = q_query.scalar()
-
-    batch_query = db.query(func.count(UploadBatch.id))
-    if user_id is not None:
-        batch_query = batch_query.filter(UploadBatch.user_id == user_id)
-    total_batches = batch_query.scalar()
-
-    total_tags = db.query(func.count(KnowledgeTag.id)).scalar()
-
-    # 按科目统计
-    subject_q = db.query(UploadBatch.subject, func.count(Question.id)).join(Question)
-    if user_id is not None:
-        subject_q = subject_q.filter(Question.user_id == user_id)
-    subject_stats = subject_q.group_by(UploadBatch.subject).all()
-
-    return {
-        "total_questions": total_questions or 0,
-        "total_batches": total_batches or 0,
-        "total_tags": total_tags or 0,
-        "by_subject": {s: c for s, c in subject_stats}
-    }
-
-
-def get_history_questions(
-    db: Session,
-    start_date: Optional[datetime] = None,
-    end_date: Optional[datetime] = None,
-    page: int = 1,
-    page_size: int = 20
-) -> Tuple[List[Question], int]:
-    """
-    分页查询历史题目（全部题目）
-
-    Args:
-        db: 数据库会话
-        start_date: 开始日期筛选
-        end_date: 结束日期筛选
-        page: 页码（从1开始）
-        page_size: 每页数量
-
-    Returns:
-        (题目列表, 总数)
-    """
-    query = db.query(Question).join(UploadBatch)
-
-    if start_date:
-        query = query.filter(UploadBatch.upload_time >= start_date)
-    if end_date:
-        from datetime import timedelta
-        query = query.filter(UploadBatch.upload_time < end_date + timedelta(days=1))
-
-    # 获取总数
-    total = query.count()
-
-    # 分页查询
-    offset = (page - 1) * page_size
-    questions = (
-        query.options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
-        .order_by(Question.created_at.desc())
-        .offset(offset)
-        .limit(page_size)
-        .all()
-    )
-
-    return questions, total
-
-
-def search_questions(
-    db: Session,
-    keyword: Optional[str] = None,
-    knowledge_tag: Optional[str] = None,
-    question_type: Optional[str] = None,
-    page: int = 1,
-    page_size: int = 20
-) -> Tuple[List[Question], int]:
-    """
-    搜索题目（知识点/题型/关键字）
-
-    Args:
-        db: 数据库会话
-        keyword: 关键字搜索（匹配题目内容 content_json）
-        knowledge_tag: 知识点标签筛选
-        question_type: 题型筛选
-        page: 页码（从1开始）
-        page_size: 每页数量
-
-    Returns:
-        (题目列表, 总数)
-    """
-    query = db.query(Question)
-
-    # 关键字搜索：匹配 content_json 中的内容
-    if keyword:
-        escaped = re.sub(r"([%_\\])", r"\\\1", keyword)
-        query = query.filter(Question.content_json.ilike(f"%{escaped}%"))
-
-    # 题型筛选
-    if question_type:
-        query = query.filter(Question.question_type == question_type)
-
-    # 知识点标签筛选（支持逗号分隔多选，OR 语义）
-    if knowledge_tag:
-        tag_list = _parse_tag_list(knowledge_tag)
-        if tag_list:
-            query = query.join(QuestionTagMapping).join(KnowledgeTag).filter(
-                KnowledgeTag.tag_name.in_(tag_list)
-            )
-
-    # 获取总数（需要先去除distinct，因为join可能产生重复）
-    total = query.distinct().count()
-
-    # 分页查询
-    offset = (page - 1) * page_size
-    questions = (
-        query.distinct()
-        .options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
-        .order_by(Question.created_at.desc())
-        .offset(offset)
-        .limit(page_size)
-        .all()
-    )
-
-    return questions, total
-
-
-def get_knowledge_stats(db: Session, subject: Optional[str] = None, limit: Optional[int] = None, user_id=None) -> List[Dict]:
-    """
-    获取知识点统计信息
-
-    Args:
-        subject: 可选，按学科筛选
-        limit: 可选，返回前 N 条
-        user_id: 可选，按用户隔离
-
-    Returns:
-        [{"tag_name": "xxx", "count": 10}, ...]
-    """
-    query = db.query(
-        KnowledgeTag.tag_name,
-        func.count(QuestionTagMapping.question_id).label("count")
-    ).join(
-        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
-    )
-
-    if user_id is not None:
-        query = query.join(Question, Question.id == QuestionTagMapping.question_id).filter(
-            Question.user_id == user_id
-        )
-
-    if subject:
-        query = query.filter(KnowledgeTag.subject == subject)
-
-    query = query.group_by(
-        KnowledgeTag.id, KnowledgeTag.tag_name
-    ).order_by(
-        func.count(QuestionTagMapping.question_id).desc()
-    )
-
-    if limit:
-        query = query.limit(limit)
-
-    stats = query.all()
-
-    return [{"tag_name": tag_name, "count": count} for tag_name, count in stats]
-
-
-def delete_question(db: Session, question_id: int) -> bool:
-    """
-    删除题目
-
-    Args:
-        db: 数据库会话
-        question_id: 题目ID
-
-    Returns:
-        是否删除成功
-    """
-    question = db.query(Question).filter(Question.id == question_id).first()
-    if not question:
-        return False
-
-    try:
-        # 删除关联的标签映射
-        db.query(QuestionTagMapping).filter(QuestionTagMapping.question_id == question_id).delete()
-
-        # 删除题目
-        db.delete(question)
-        db.commit()
-    except Exception as e:
-        db.rollback()
-        logger.error(f"删除题目 {question_id} 失败: {e}")
-        raise
-
-    return True
-
-
-def query_questions(
-    db: Session,
-    subject: Optional[str] = None,
-    knowledge_tag: Optional[str] = None,
-    question_type: Optional[str] = None,
-    keyword: Optional[str] = None,
-    start_date: Optional[datetime] = None,
-    end_date: Optional[datetime] = None,
-    review_status: Optional[str] = None,
-    page: int = 1,
-    page_size: int = 20,
-    user_id=None,
-) -> Tuple[List[Question], int]:
-    """
-    统一查询题目（合并 get_history_questions 和 search_questions 的能力）
-
-    支持所有筛选条件任意组合。
-    """
-    query = db.query(Question).join(UploadBatch)
-
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-
-    if subject:
-        query = query.filter(UploadBatch.subject == subject)
-
-    if question_type:
-        query = query.filter(Question.question_type == question_type)
-
-    if keyword:
-        escaped = re.sub(r"([%_\\])", r"\\\1", keyword)
-        query = query.filter(Question.content_json.ilike(f"%{escaped}%"))
-
-    if knowledge_tag:
-        tag_list = _parse_tag_list(knowledge_tag)
-        if tag_list:
-            query = query.join(QuestionTagMapping).join(KnowledgeTag).filter(
-                KnowledgeTag.tag_name.in_(tag_list)
-            )
-
-    if start_date:
-        query = query.filter(Question.created_at >= start_date)
-    if end_date:
-        from datetime import timedelta
-        query = query.filter(Question.created_at < end_date + timedelta(days=1))
-
-    if review_status:
-        query = query.filter(Question.review_status == review_status)
-
-    total = query.distinct().count()
-
-    offset = (page - 1) * page_size
-    questions = (
-        query.distinct()
-        .options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
-        .order_by(Question.created_at.desc())
-        .offset(offset)
-        .limit(page_size)
-        .all()
-    )
-
-    return questions, total
-
-
-def update_user_answer(db: Session, question_id: int, user_answer: str) -> Optional[Question]:
-    """更新用户答案"""
-    question = db.query(Question).filter(Question.id == question_id).first()
-    if not question:
-        return None
-
-    try:
-        question.user_answer = user_answer
-        question.updated_at = datetime.utcnow()
-        db.commit()
-        db.refresh(question)
-        return question
-    except Exception as e:
-        db.rollback()
-        logger.error(f"更新题目 {question_id} 答案失败: {e}")
-        raise
-
-
-def get_existing_question_types(db: Session, user_id=None) -> List[str]:
-    """获取数据库中已有的所有题型（去重）"""
-    query = db.query(Question.question_type).distinct().filter(
-        Question.question_type.isnot(None),
-        Question.question_type != "",
-    )
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-    return [r[0] for r in query.all()]
-
-
-def get_questions_by_ids(db: Session, question_ids: List[int]) -> List[Question]:
-    """按 ID 列表批量查询题目"""
-    if not question_ids:
-        return []
-    return (
-        db.query(Question)
-        .options(joinedload(Question.batch), joinedload(Question.tags).joinedload(QuestionTagMapping.tag))
-        .filter(Question.id.in_(question_ids))
-        .all()
-    )
-
-
-VALID_REVIEW_STATUSES = ('待复习', '复习中', '已掌握')
-
-
-def update_review_status(db: Session, question_id: int, review_status: str) -> Optional[Question]:
-    """更新题目复习状态"""
-    if review_status not in VALID_REVIEW_STATUSES:
-        raise ValueError(f"无效的复习状态: {review_status}，可选值: {VALID_REVIEW_STATUSES}")
-
-    question = db.query(Question).filter(Question.id == question_id).first()
-    if not question:
-        return None
-
-    try:
-        question.review_status = review_status
-        question.updated_at = datetime.utcnow()
-        db.commit()
-        db.refresh(question)
-        return question
-    except Exception as e:
-        db.rollback()
-        logger.error(f"更新题目 {question_id} 复习状态失败: {e}")
-        raise
-
-
-def get_review_status_stats(db: Session, subject: Optional[str] = None, user_id=None) -> Dict[str, int]:
-    """按复习状态分组统计数量"""
-    query = _filter_by_subject(db.query(
-        Question.review_status,
-        func.count(Question.id)
-    ), subject)
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-
-    rows = query.group_by(Question.review_status).all()
-
-    result = {s: 0 for s in VALID_REVIEW_STATUSES}
-    for status, count in rows:
-        key = status or '待复习'
-        if key in result:
-            result[key] += count
-        else:
-            result['待复习'] += count
-    return result
-
-
-def get_daily_counts(db: Session, days: int = 7, subject: Optional[str] = None, user_id=None) -> List[Dict[str, Any]]:
-    """获取最近 N 天每日新增题目数 + 每日新增已掌握数"""
-    from datetime import timedelta
-    cutoff = datetime.utcnow().replace(hour=0, minute=0, second=0, microsecond=0) - timedelta(days=days - 1)
-
-    # SQLite 兼容：使用 func.date() 提取日期字符串
-    date_expr = func.date(Question.created_at)
-
-    # 新增题目数
-    query = _filter_by_subject(db.query(
-        date_expr.label('date'),
-        func.count(Question.id).label('count')
-    ).filter(Question.created_at >= cutoff), subject)
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-    rows = query.group_by(date_expr).order_by(date_expr).all()
-    date_map = {str(row.date): row.count for row in rows}
-
-    # 每日新增已掌握数（按 updated_at 统计）
-    mastered_date_expr = func.date(Question.updated_at)
-    mq = _filter_by_subject(db.query(
-        mastered_date_expr.label('date'),
-        func.count(Question.id).label('count')
-    ).filter(
-        Question.updated_at >= cutoff,
-        Question.review_status == '已掌握',
-    ), subject)
-    if user_id is not None:
-        mq = mq.filter(Question.user_id == user_id)
-    mastered_rows = mq.group_by(mastered_date_expr).order_by(mastered_date_expr).all()
-    mastered_map = {str(row.date): row.count for row in mastered_rows}
-
-    # 填充缺失的日期
-    result = []
-    for i in range(days):
-        d = cutoff + timedelta(days=i)
-        date_key = d.strftime('%Y-%m-%d')
-        date_str = d.strftime('%m-%d')
-        result.append({
-            'date': date_str,
-            'count': date_map.get(date_key, 0),
-            'mastered': mastered_map.get(date_key, 0),
-        })
-
-    return result
-
-
-def get_tag_status_stats(db: Session, subject: Optional[str] = None, limit: int = 10, user_id=None) -> List[Dict]:
-    """
-    知识点 × 掌握状态统计（堆叠柱状图用）
-
-    Returns:
-        [{"tag_name": "函数", "待复习": 3, "复习中": 2, "已掌握": 5}, ...]
-    """
-    query = db.query(
-        KnowledgeTag.tag_name,
-        Question.review_status,
-        func.count(Question.id).label('count')
-    ).join(
-        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
-    ).join(
-        Question, Question.id == QuestionTagMapping.question_id
-    )
-
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-    if subject:
-        query = query.filter(KnowledgeTag.subject == subject)
-
-    rows = query.group_by(
-        KnowledgeTag.tag_name, Question.review_status
-    ).all()
-
-    # 按 tag 聚合
-    tag_data = {}
-    tag_totals = {}
-    for tag_name, status, count in rows:
-        if tag_name not in tag_data:
-            tag_data[tag_name] = {'tag_name': tag_name, **{s: 0 for s in VALID_REVIEW_STATUSES}}
-            tag_totals[tag_name] = 0
-        key = status or '待复习'
-        if key in VALID_REVIEW_STATUSES:
-            tag_data[tag_name][key] += count
-        else:
-            tag_data[tag_name]['待复习'] += count
-        tag_totals[tag_name] += count
-
-    # 按总数降序，取 top N
-    sorted_tags = sorted(tag_data.keys(), key=lambda t: tag_totals[t], reverse=True)[:limit]
-    return [tag_data[t] for t in sorted_tags]
-
-
-def get_tag_type_stats(db: Session, subject: Optional[str] = None, tag_limit: int = 8, user_id=None) -> Dict[str, Any]:
-    """
-    知识点 × 题型交叉统计（热力图用）
-
-    Returns:
-        {"tags": ["函数", ...], "types": ["选择题", ...], "data": [[3, 1, ...], ...]}
-    """
-    query = db.query(
-        KnowledgeTag.tag_name,
-        Question.question_type,
-        func.count(Question.id).label('count')
-    ).join(
-        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
-    ).join(
-        Question, Question.id == QuestionTagMapping.question_id
-    ).filter(
-        Question.question_type.isnot(None),
-        Question.question_type != '',
-    )
-
-    if user_id is not None:
-        query = query.filter(Question.user_id == user_id)
-    if subject:
-        query = query.filter(KnowledgeTag.subject == subject)
-
-    rows = query.group_by(
-        KnowledgeTag.tag_name, Question.question_type
-    ).all()
-
-    # 收集所有 tag 和 type
-    tag_totals = {}
-    type_set = set()
-    cross = {}
-    for tag_name, q_type, count in rows:
-        tag_totals[tag_name] = tag_totals.get(tag_name, 0) + count
-        type_set.add(q_type)
-        cross[(tag_name, q_type)] = count
-
-    # 按总数取 top N tag
-    sorted_tags = sorted(tag_totals.keys(), key=lambda t: tag_totals[t], reverse=True)[:tag_limit]
-    sorted_types = sorted(type_set)
-
-    # 构建矩阵
-    data = []
-    for tag in sorted_tags:
-        row = [cross.get((tag, t), 0) for t in sorted_types]
-        data.append(row)
-
-    return {"tags": sorted_tags, "types": sorted_types, "data": data}
-
-
-# ============================================================
-# 教学辅导对话 CRUD
-# ============================================================
-
-
-def update_question_answer(db: Session, question_id: int, answer: str) -> Optional[Question]:
-    """保存/更新题目答案（Markdown 格式）"""
-    question = db.query(Question).filter(Question.id == question_id).first()
-    if not question:
-        return None
-
-    try:
-        question.answer = answer
-        question.updated_at = datetime.utcnow()
-        db.commit()
-        db.refresh(question)
-        return question
-    except Exception as e:
-        db.rollback()
-        logger.error(f"保存题目 {question_id} 答案失败: {e}")
-        raise
-
-
-def create_chat_session(db: Session, question_id: int) -> ChatSession:
-    """为题目创建新的对话会话"""
-    session = ChatSession(question_id=question_id)
-    db.add(session)
-    try:
-        db.commit()
-        db.refresh(session)
-        return session
-    except Exception as e:
-        db.rollback()
-        logger.error(f"创建对话会话失败: {e}")
-        raise
-
-
-_VALID_ROLES = ('user', 'assistant')
-
-
-def add_chat_message(db: Session, session_id: int, role: str, content: str) -> ChatMessage:
-    """向对话中追加一条消息"""
-    if role not in _VALID_ROLES:
-        raise ValueError(f"无效的消息角色: {role}（可选: {', '.join(_VALID_ROLES)}）")
-    msg = ChatMessage(session_id=session_id, role=role, content=content)
-    db.add(msg)
-    try:
-        # 同步更新会话的 updated_at（直接 UPDATE 避免加载整行）
-        db.query(ChatSession).filter(ChatSession.id == session_id).update(
-            {"updated_at": datetime.utcnow()}, synchronize_session=False
-        )
-        db.commit()
-        db.refresh(msg)
-        return msg
-    except Exception as e:
-        db.rollback()
-        logger.error(f"添加对话消息失败: {e}")
-        raise
-
-
-def get_chat_messages(
-    db: Session,
-    session_id: int,
-    limit: int = 30,
-    before_id: Optional[int] = None,
-) -> Dict[str, Any]:
-    """
-    游标分页获取对话消息
-
-    Args:
-        session_id: 会话 ID
-        limit: 每次返回的消息数
-        before_id: 游标，返回 ID 小于此值的消息
-
-    Returns:
-        {"messages": [...], "hasMore": bool}
-    """
-    query = db.query(ChatMessage).filter(ChatMessage.session_id == session_id)
-    if before_id:
-        query = query.filter(ChatMessage.id < before_id)
-
-    # 按 ID 降序取 limit+1 条，判断是否还有更早消息
-    rows = query.order_by(ChatMessage.id.desc()).limit(limit + 1).all()
-
-    has_more = len(rows) > limit
-    messages = rows[:limit]
-    messages.reverse()  # 恢复正序
-
-    return {
-        "messages": [
-            {"id": m.id, "role": m.role, "content": m.content, "created_at": m.created_at.isoformat() if m.created_at else None}
-            for m in messages
-        ],
-        "hasMore": has_more,
-    }
-
-
-def get_chat_sessions_by_question(db: Session, question_id: int, limit: int = 50) -> List[ChatSession]:
-    """获取某道题目的对话会话（按更新时间降序，默认最多 50 条）"""
-    return (
-        db.query(ChatSession)
-        .filter(ChatSession.question_id == question_id)
-        .order_by(ChatSession.updated_at.desc())
-        .limit(limit)
-        .all()
-    )
-
-
-# ============================================================
-# 分割历史记录 CRUD
-# ============================================================
-
-
-MAX_SPLIT_RECORDS = 20
-
-
-def save_split_record(
-    db: Session,
-    subject: Optional[str],
-    model_provider: str,
-    file_names: List[str],
-    questions: List[Dict],
-    user_id=None,
-) -> SplitRecord:
-    """保存一次分割操作的完整结果，超过上限自动清理最旧记录"""
-    record = SplitRecord(
-        user_id=user_id,
-        subject=subject,
-        model_provider=model_provider,
-        file_names_json=json.dumps(file_names, ensure_ascii=False),
-        questions_json=json.dumps(questions, ensure_ascii=False),
-        question_count=len(questions),
-    )
-    db.add(record)
-    try:
-        db.commit()
-        db.refresh(record)
-        _cleanup_old_split_records(db)
-        return record
-    except Exception as e:
-        db.rollback()
-        logger.error(f"保存分割记录失败: {e}")
-        raise
-
-
-def _cleanup_old_split_records(db: Session):
-    """删除超出上限的最旧分割记录"""
-    count = db.query(func.count(SplitRecord.id)).scalar()
-    if count <= MAX_SPLIT_RECORDS:
-        return
-    overflow = count - MAX_SPLIT_RECORDS
-    old_ids = [
-        row[0] for row in
-        db.query(SplitRecord.id)
-        .order_by(SplitRecord.created_at.asc())
-        .limit(overflow)
-        .all()
-    ]
-    try:
-        db.query(SplitRecord).filter(SplitRecord.id.in_(old_ids)).delete(synchronize_session=False)
-        db.commit()
-        logger.info(f"已清理 {overflow} 条过期分割记录")
-    except Exception as e:
-        db.rollback()
-        logger.error(f"清理分割记录失败: {e}")
-
-
-def get_recent_split_records(db, limit: int = 10, user_id=None):
-    """获取最近 N 条分割记录（不加载 questions_json 大字段）"""
-    from sqlalchemy.orm import defer
-    return (
-        db.query(SplitRecord)
-        .options(defer(SplitRecord.questions_json))
-        .order_by(SplitRecord.created_at.desc())
-        .limit(limit)
-        .all()
-    )
-
-
-def get_split_record_by_id(db: Session, record_id: int) -> Optional[SplitRecord]:
-    """按 ID 获取单条分割记录（含完整 questions_json）"""
-    return db.query(SplitRecord).filter(SplitRecord.id == record_id).first()
-
-
-def get_all_chat_sessions(
-    db: Session,
-    page: int = 1,
-    page_size: int = 20,
-) -> Tuple[List[ChatSession], int]:
-    """分页获取所有对话会话"""
-    query = db.query(ChatSession)
-    total = query.count()
-    offset = (page - 1) * page_size
-    sessions = (
-        query.options(selectinload(ChatSession.question))
-        .order_by(ChatSession.updated_at.desc())
-        .offset(offset)
-        .limit(page_size)
-        .all()
-    )
-    return sessions, total
-
-
-# ── Provider Config CRUD ──────────────────────────────────────
-
-
-def _mask_secret(value: str, visible: int = 4) -> str:
-    """脱敏显示密钥"""
-    if not value:
-        return ""
-    if len(value) <= visible:
-        return "*" * len(value)
-    return "*" * 4 + value[-visible:]
-
-
-def _serialize_provider(p: ProviderConfig) -> dict:
-    """序列化 ProviderConfig 为前端可用 dict（密钥脱敏）"""
-    d = {
-        "id": p.id,
-        "name": p.name,
-        "is_active": p.is_active,
-        "api_key_set": bool(p.api_key),
-        "api_key_hint": _mask_secret(p.api_key),
-        "base_url": p.base_url or "",
-        "model_name": p.model_name or "",
-    }
-    if p.category == "openai":
-        d["light_model_name"] = p.light_model_name or ""
-        d["supports_function_calling"] = p.supports_function_calling
-    elif p.category == "paddleocr":
-        d["use_doc_orientation"] = p.use_doc_orientation
-        d["use_doc_unwarping"] = p.use_doc_unwarping
-        d["use_chart_recognition"] = p.use_chart_recognition
-    return d
-
-
-def get_user_providers(db: Session, user_id: int) -> dict:
-    """获取用户的所有 provider 配置，按类别分组返回"""
-    providers = db.query(ProviderConfig).filter(
-        ProviderConfig.user_id == user_id
-    ).order_by(ProviderConfig.created_at).all()
-
-    result = {
-        "openai_providers": [],
-        "anthropic_providers": [],
-        "paddleocr_providers": [],
-        "active_openai_id": None,
-        "active_anthropic_id": None,
-        "active_paddleocr_id": None,
-    }
-    category_key = {
-        "openai": "openai_providers",
-        "anthropic": "anthropic_providers",
-        "paddleocr": "paddleocr_providers",
-    }
-    active_key = {
-        "openai": "active_openai_id",
-        "anthropic": "active_anthropic_id",
-        "paddleocr": "active_paddleocr_id",
-    }
-
-    for p in providers:
-        key = category_key.get(p.category)
-        if key:
-            result[key].append(_serialize_provider(p))
-        if p.is_active:
-            ak = active_key.get(p.category)
-            if ak:
-                result[ak] = p.id
-
-    return result
-
-
-def save_user_providers(db: Session, user_id: int, data: dict) -> None:
-    """保存用户的 provider 配置（全量覆盖）
-
-    data 结构:
-        openai_providers: [{ id, name, api_key?, base_url, model_name, ... }]
-        anthropic_providers: [...]
-        paddleocr_providers: [...]
-        active_openai_id: str | None
-        active_anthropic_id: str | None
-        active_paddleocr_id: str | None
-    """
-    active_ids = {
-        "openai": data.get("active_openai_id"),
-        "anthropic": data.get("active_anthropic_id"),
-        "paddleocr": data.get("active_paddleocr_id"),
-    }
-
-    # 读取已有配置（用于保留未重新提交的 api_key）
-    existing = {
-        p.id: p for p in db.query(ProviderConfig).filter(
-            ProviderConfig.user_id == user_id
-        ).all()
-    }
-
-    # 收集本次提交的所有 ID
-    submitted_ids = set()
-    items_to_save = []
-
-    category_map = {
-        "openai_providers": "openai",
-        "anthropic_providers": "anthropic",
-        "paddleocr_providers": "paddleocr",
-    }
-
-    for list_key, category in category_map.items():
-        for item in data.get(list_key, []):
-            pid = item.get("id")
-            submitted_ids.add(pid)
-            old = existing.get(pid)
-
-            # API Key：前端提交了新值则更新，否则保留旧值
-            new_api_key = item.get("api_key") or item.get("api_token") or ""
-            if not new_api_key and old:
-                new_api_key = old.api_key or ""
-
-            base_url = item.get("base_url") or item.get("api_url") or ""
-            model_name = item.get("model_name") or item.get("model") or ""
-
-            items_to_save.append({
-                "id": pid,
-                "category": category,
-                "name": item.get("name", ""),
-                "is_active": pid == active_ids.get(category),
-                "api_key": new_api_key,
-                "base_url": base_url,
-                "model_name": model_name,
-                "light_model_name": item.get("light_model_name", ""),
-                "supports_function_calling": item.get("supports_function_calling", True),
-                "use_doc_orientation": item.get("use_doc_orientation", False),
-                "use_doc_unwarping": item.get("use_doc_unwarping", False),
-                "use_chart_recognition": item.get("use_chart_recognition", False),
-            })
-
-    # 删除已移除的 provider
-    for pid in set(existing.keys()) - submitted_ids:
-        db.delete(existing[pid])
-
-    # 更新或新增
-    for item in items_to_save:
-        old = existing.get(item["id"])
-        if old:
-            for k, v in item.items():
-                setattr(old, k, v)
-        else:
-            p = ProviderConfig(user_id=user_id, **item)
-            db.add(p)
-
-    db.commit()
-
-
-def get_active_provider(db: Session, user_id: int, category: str) -> Optional[ProviderConfig]:
-    """获取用户指定类别的激活 provider（用于后端调用 LLM/OCR 时读取凭据）"""
-    return db.query(ProviderConfig).filter(
-        ProviderConfig.user_id == user_id,
-        ProviderConfig.category == category,
-        ProviderConfig.is_active == True,
-    ).first()
diff --git a/backend/db/crud/__init__.py b/backend/db/crud/__init__.py
new file mode 100644
index 00000000..bd8967e2
--- /dev/null
+++ b/backend/db/crud/__init__.py
@@ -0,0 +1,187 @@
+"""
+数据库 CRUD 操作
+
+所有公共函数从子模块汇总导出，保持 ``from db import crud; crud.xxx()`` 的使用方式不变。
+"""
+
+from typing import Optional
+
+from db.models import UploadBatch, Question
+
+
+# ── 共享过滤辅助函数 ──────────────────────────────────────────
+
+def _filter_by_subject(query, subject: Optional[str]):
+    """如有学科筛选，为 Question 查询追加 JOIN UploadBatch 过滤"""
+    if subject:
+        return query.join(UploadBatch).filter(UploadBatch.subject == subject)
+    return query
+
+
+def _filter_by_user(query, user_id):
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return query
+
+
+# ── 子模块导出 ────────────────────────────────────────────────
+
+from db.crud.users import (
+    create_user,
+    get_user_by_email,
+    get_user_by_id,
+    get_user_by_login,
+    update_user_password,
+)
+
+from db.crud.email_verification import (
+    get_verification_by_email,
+    upsert_registration_code,
+    delete_verification_by_email,
+    increment_verification_attempts,
+)
+
+from db.crud.tags import (
+    _parse_tag_list,
+    get_or_create_tag,
+    get_existing_tag_names,
+    get_all_tags,
+)
+
+from db.crud.questions import (
+    compute_content_hash,
+    question_exists,
+    save_questions_to_db,
+    get_questions_by_subject,
+    get_questions_by_tag,
+    get_history_questions,
+    search_questions,
+    query_questions,
+    get_questions_by_ids,
+    delete_question,
+    update_user_answer,
+    update_question_answer,
+    update_review_status,
+    get_existing_subjects,
+    get_existing_question_types,
+    VALID_REVIEW_STATUSES,
+)
+
+from db.crud.stats import (
+    get_statistics,
+    get_knowledge_stats,
+    get_review_status_stats,
+    get_today_mastered_count,
+    get_daily_counts,
+    get_tag_status_stats,
+    get_tag_type_stats,
+)
+
+from db.crud.chat import (
+    create_chat_session,
+    add_chat_message,
+    get_chat_messages,
+    get_chat_sessions_by_question,
+    get_all_chat_sessions,
+    get_user_chat_sessions,
+    update_chat_session_title,
+    delete_chat_session,
+)
+
+from db.crud.split_records import (
+    MAX_SPLIT_RECORDS,
+    save_split_record,
+    _cleanup_old_split_records,
+    get_recent_split_records,
+    get_split_record_by_id,
+)
+
+from db.crud.providers import (
+    _mask_secret,
+    _serialize_provider,
+    get_user_providers,
+    save_user_providers,
+    get_active_provider,
+)
+
+from db.crud.notes import (
+    save_note,
+    get_notes,
+    get_note_by_id,
+    update_note,
+    delete_note,
+)
+
+__all__ = [
+    # shared helpers
+    "_filter_by_subject",
+    "_filter_by_user",
+    # users
+    "create_user",
+    "get_user_by_email",
+    "get_user_by_id",
+    "get_user_by_login",
+    "update_user_password",
+    # email verification
+    "get_verification_by_email",
+    "upsert_registration_code",
+    "delete_verification_by_email",
+    "increment_verification_attempts",
+    # tags
+    "_parse_tag_list",
+    "get_or_create_tag",
+    "get_existing_tag_names",
+    "get_all_tags",
+    # questions
+    "compute_content_hash",
+    "question_exists",
+    "save_questions_to_db",
+    "get_questions_by_subject",
+    "get_questions_by_tag",
+    "get_history_questions",
+    "search_questions",
+    "query_questions",
+    "get_questions_by_ids",
+    "delete_question",
+    "update_user_answer",
+    "update_question_answer",
+    "update_review_status",
+    "get_existing_subjects",
+    "get_existing_question_types",
+    "VALID_REVIEW_STATUSES",
+    # stats
+    "get_statistics",
+    "get_knowledge_stats",
+    "get_review_status_stats",
+    "get_today_mastered_count",
+    "get_daily_counts",
+    "get_tag_status_stats",
+    "get_tag_type_stats",
+    # chat
+    "create_chat_session",
+    "add_chat_message",
+    "get_chat_messages",
+    "get_chat_sessions_by_question",
+    "get_all_chat_sessions",
+    "get_user_chat_sessions",
+    "update_chat_session_title",
+    "delete_chat_session",
+    # split_records
+    "MAX_SPLIT_RECORDS",
+    "save_split_record",
+    "_cleanup_old_split_records",
+    "get_recent_split_records",
+    "get_split_record_by_id",
+    # providers
+    "_mask_secret",
+    "_serialize_provider",
+    "get_user_providers",
+    "save_user_providers",
+    "get_active_provider",
+    # notes
+    "save_note",
+    "get_notes",
+    "get_note_by_id",
+    "update_note",
+    "delete_note",
+]
diff --git a/backend/db/crud/chat.py b/backend/db/crud/chat.py
new file mode 100644
index 00000000..83fc3556
--- /dev/null
+++ b/backend/db/crud/chat.py
@@ -0,0 +1,176 @@
+"""对话 CRUD（支持题目绑定对话和独立对话）"""
+
+import logging
+from datetime import datetime
+from typing import List, Dict, Any, Optional, Tuple
+
+from sqlalchemy.orm import Session, selectinload
+
+from db.models import ChatSession, ChatMessage
+
+logger = logging.getLogger(__name__)
+
+
+def create_chat_session(
+    db: Session,
+    question_id: int = None,
+    user_id: int = None,
+    title: str = "新对话",
+) -> ChatSession:
+    """创建对话会话
+
+    Args:
+        question_id: 绑定的题目 ID（可选，None 为独立对话）
+        user_id: 用户 ID
+        title: 对话标题
+    """
+    session = ChatSession(question_id=question_id, user_id=user_id, title=title)
+    db.add(session)
+    try:
+        db.commit()
+        db.refresh(session)
+        return session
+    except Exception as e:
+        db.rollback()
+        logger.error(f"创建对话会话失败: {e}")
+        raise
+
+
+_VALID_ROLES = ('user', 'assistant')
+
+
+def add_chat_message(db: Session, session_id: int, role: str, content: str, user_id=None) -> ChatMessage:
+    """向对话中追加一条消息"""
+    if role not in _VALID_ROLES:
+        raise ValueError(f"无效的消息角色: {role}（可选: {', '.join(_VALID_ROLES)}）")
+    # 验证 session 归属
+    if user_id is not None:
+        owner_check = db.query(ChatSession).filter(
+            ChatSession.id == session_id, ChatSession.user_id == user_id
+        ).first()
+        if not owner_check:
+            raise ValueError("对话不存在或无权操作")
+    msg = ChatMessage(session_id=session_id, role=role, content=content)
+    db.add(msg)
+    try:
+        db.query(ChatSession).filter(ChatSession.id == session_id).update(
+            {"updated_at": datetime.utcnow()}, synchronize_session=False
+        )
+        db.commit()
+        db.refresh(msg)
+        return msg
+    except Exception as e:
+        db.rollback()
+        logger.error(f"添加对话消息失败: {e}")
+        raise
+
+
+def get_chat_messages(
+    db: Session,
+    session_id: int,
+    limit: int = 30,
+    before_id: Optional[int] = None,
+    user_id=None,
+) -> Dict[str, Any]:
+    """游标分页获取对话消息"""
+    # 验证 session 归属
+    if user_id is not None:
+        owner_check = db.query(ChatSession).filter(
+            ChatSession.id == session_id, ChatSession.user_id == user_id
+        ).first()
+        if not owner_check:
+            return {"messages": [], "hasMore": False}
+    query = db.query(ChatMessage).filter(ChatMessage.session_id == session_id)
+    if before_id:
+        query = query.filter(ChatMessage.id < before_id)
+
+    rows = query.order_by(ChatMessage.id.desc()).limit(limit + 1).all()
+    has_more = len(rows) > limit
+    messages = rows[:limit]
+    messages.reverse()
+
+    return {
+        "messages": [
+            {"id": m.id, "role": m.role, "content": m.content, "created_at": m.created_at.isoformat() if m.created_at else None}
+            for m in messages
+        ],
+        "hasMore": has_more,
+    }
+
+
+def get_chat_sessions_by_question(db: Session, question_id: int, limit: int = 50, user_id=None) -> List[ChatSession]:
+    """获取某道题目的对话会话"""
+    query = db.query(ChatSession).filter(ChatSession.question_id == question_id)
+    if user_id is not None:
+        query = query.filter(ChatSession.user_id == user_id)
+    return query.order_by(ChatSession.updated_at.desc()).limit(limit).all()
+
+
+def get_user_chat_sessions(
+    db: Session,
+    user_id: int,
+    page: int = 1,
+    page_size: int = 20,
+) -> Tuple[List[ChatSession], int]:
+    """获取用户的独立对话列表（不含题目绑定的对话）"""
+    query = db.query(ChatSession).filter(
+        ChatSession.user_id == user_id,
+        ChatSession.question_id == None,
+    )
+    total = query.count()
+    sessions = (
+        query.order_by(ChatSession.updated_at.desc())
+        .offset((page - 1) * page_size)
+        .limit(page_size)
+        .all()
+    )
+    return sessions, total
+
+
+def get_all_chat_sessions(
+    db: Session,
+    page: int = 1,
+    page_size: int = 20,
+    user_id=None,
+) -> Tuple[List[ChatSession], int]:
+    """分页获取对话会话（user_id 非 None 时按用户过滤）"""
+    query = db.query(ChatSession)
+    if user_id is not None:
+        query = query.filter(ChatSession.user_id == user_id)
+    total = query.count()
+    sessions = (
+        query.options(selectinload(ChatSession.question))
+        .order_by(ChatSession.updated_at.desc())
+        .offset((page - 1) * page_size)
+        .limit(page_size)
+        .all()
+    )
+    return sessions, total
+
+
+def update_chat_session_title(db: Session, session_id: int, title: str, user_id=None) -> Optional[ChatSession]:
+    """更新对话标题"""
+    query = db.query(ChatSession).filter(ChatSession.id == session_id)
+    if user_id is not None:
+        query = query.filter(ChatSession.user_id == user_id)
+    session = query.first()
+    if not session:
+        return None
+    session.title = title
+    db.commit()
+    db.refresh(session)
+    return session
+
+
+def delete_chat_session(db: Session, session_id: int, user_id=None) -> bool:
+    """删除对话（级联删除消息）"""
+    query = db.query(ChatSession).filter(ChatSession.id == session_id)
+    if user_id is not None:
+        query = query.filter(ChatSession.user_id == user_id)
+    session = query.first()
+    if not session:
+        return False
+    db.query(ChatMessage).filter(ChatMessage.session_id == session_id).delete()
+    db.delete(session)
+    db.commit()
+    return True
diff --git a/backend/db/crud/email_verification.py b/backend/db/crud/email_verification.py
new file mode 100644
index 00000000..b5d845b7
--- /dev/null
+++ b/backend/db/crud/email_verification.py
@@ -0,0 +1,52 @@
+"""注册邮箱验证码 CRUD"""
+
+from datetime import datetime
+
+from db.models import EmailVerification
+
+
+def get_verification_by_email(db, email: str):
+    return db.query(EmailVerification).filter(EmailVerification.email == email).first()
+
+
+def upsert_registration_code(
+    db,
+    email: str,
+    code_hash: str,
+    expires_at: datetime,
+    last_sent_at: datetime,
+) -> EmailVerification:
+    row = get_verification_by_email(db, email)
+    if row:
+        row.code_hash = code_hash
+        row.expires_at = expires_at
+        row.last_sent_at = last_sent_at
+        # 不重置 attempts：防止攻击者反复请求新验证码来绕过尝试次数限制
+    else:
+        row = EmailVerification(
+            email=email,
+            code_hash=code_hash,
+            expires_at=expires_at,
+            last_sent_at=last_sent_at,
+            attempts=0,
+        )
+        db.add(row)
+    db.commit()
+    db.refresh(row)
+    return row
+
+
+def delete_verification_by_email(db, email: str) -> None:
+    row = get_verification_by_email(db, email)
+    if row:
+        db.delete(row)
+        db.commit()
+
+
+def increment_verification_attempts(db, email: str) -> int:
+    row = get_verification_by_email(db, email)
+    if not row:
+        return 0
+    row.attempts = (row.attempts or 0) + 1
+    db.commit()
+    return row.attempts
diff --git a/backend/db/crud/notes.py b/backend/db/crud/notes.py
new file mode 100644
index 00000000..89b0bf9b
--- /dev/null
+++ b/backend/db/crud/notes.py
@@ -0,0 +1,180 @@
+"""
+笔记 CRUD 操作
+"""
+
+import json
+import logging
+from typing import List, Dict, Any, Optional
+
+from sqlalchemy.orm import Session, selectinload
+
+from db.models import Note, NoteTagMapping, KnowledgeTag
+
+logger = logging.getLogger(__name__)
+
+
+def save_note(
+    db: Session,
+    title: str,
+    subject: str,
+    content_markdown: str,
+    source_images: List[str],
+    ocr_text: str,
+    knowledge_tags: List[str] = None,
+    user_id: int = None,
+) -> Note:
+    """保存一条笔记到数据库
+
+    Args:
+        title: 笔记标题
+        subject: 科目
+        content_markdown: LLM 整理后的 Markdown 内容
+        source_images: 原始上传图片路径列表
+        ocr_text: OCR 识别的原始文本
+        knowledge_tags: 知识点标签名称列表
+        user_id: 用户 ID
+
+    Returns:
+        创建的 Note 对象
+    """
+    note = Note(
+        user_id=user_id,
+        title=title,
+        subject=subject,
+        content_markdown=content_markdown,
+        source_images_json=json.dumps(source_images, ensure_ascii=False),
+        ocr_text=ocr_text,
+    )
+    db.add(note)
+    db.flush()  # 获取 note.id
+
+    # 创建知识点标签关联
+    if knowledge_tags:
+        from db.crud.tags import get_or_create_tag
+        for tag_name in knowledge_tags:
+            tag_name = tag_name.strip()
+            if not tag_name:
+                continue
+            tag = get_or_create_tag(db, tag_name, subject or "")
+            mapping = NoteTagMapping(note_id=note.id, tag_id=tag.id)
+            db.add(mapping)
+
+    db.commit()
+    db.refresh(note)
+    return note
+
+
+def get_notes(
+    db: Session,
+    user_id: int = None,
+    subject: str = None,
+    knowledge_tag: str = None,
+    keyword: str = None,
+    page: int = 1,
+    page_size: int = 20,
+) -> tuple:
+    """分页查询笔记列表
+
+    Returns:
+        (笔记列表, 总数)
+    """
+    query = db.query(Note).options(
+        selectinload(Note.tags).selectinload(NoteTagMapping.tag),
+    )
+
+    if user_id is not None:
+        query = query.filter(Note.user_id == user_id)
+    if subject:
+        query = query.filter(Note.subject == subject)
+    if keyword:
+        query = query.filter(
+            Note.title.ilike(f"%{keyword}%") |
+            Note.content_markdown.ilike(f"%{keyword}%")
+        )
+    if knowledge_tag:
+        query = query.join(NoteTagMapping).join(KnowledgeTag).filter(
+            KnowledgeTag.tag_name == knowledge_tag
+        )
+
+    total = query.count()
+    notes = (
+        query.order_by(Note.updated_at.desc())
+        .offset((page - 1) * page_size)
+        .limit(page_size)
+        .all()
+    )
+    return notes, total
+
+
+def get_note_by_id(db: Session, note_id: int, user_id=None) -> Optional[Note]:
+    """根据 ID 获取单条笔记"""
+    query = db.query(Note).options(
+        selectinload(Note.tags).selectinload(NoteTagMapping.tag),
+    ).filter(Note.id == note_id)
+    if user_id is not None:
+        query = query.filter(Note.user_id == user_id)
+    return query.first()
+
+
+def update_note(
+    db: Session,
+    note_id: int,
+    title: str = None,
+    content_markdown: str = None,
+    subject: str = None,
+    knowledge_tags: List[str] = None,
+    user_id=None,
+) -> Optional[Note]:
+    """更新笔记内容
+
+    Args:
+        note_id: 笔记 ID
+        title: 新标题（None 不更新）
+        content_markdown: 新 Markdown 内容（None 不更新）
+        subject: 新科目（None 不更新）
+        knowledge_tags: 新标签列表（None 不更新，空列表清空标签）
+
+    Returns:
+        更新后的 Note 对象，不存在返回 None
+    """
+    query = db.query(Note).filter(Note.id == note_id)
+    if user_id is not None:
+        query = query.filter(Note.user_id == user_id)
+    note = query.first()
+    if not note:
+        return None
+
+    if title is not None:
+        note.title = title
+    if content_markdown is not None:
+        note.content_markdown = content_markdown
+    if subject is not None:
+        note.subject = subject
+
+    # 更新标签：先删后建
+    if knowledge_tags is not None:
+        db.query(NoteTagMapping).filter(NoteTagMapping.note_id == note_id).delete()
+        from db.crud.tags import get_or_create_tag
+        for tag_name in knowledge_tags:
+            tag_name = tag_name.strip()
+            if not tag_name:
+                continue
+            tag = get_or_create_tag(db, tag_name, note.subject or "")
+            db.add(NoteTagMapping(note_id=note.id, tag_id=tag.id))
+
+    db.commit()
+    db.refresh(note)
+    return note
+
+
+def delete_note(db: Session, note_id: int, user_id=None) -> bool:
+    """删除笔记（级联删除标签关联）"""
+    query = db.query(Note).filter(Note.id == note_id)
+    if user_id is not None:
+        query = query.filter(Note.user_id == user_id)
+    note = query.first()
+    if not note:
+        return False
+    db.delete(note)
+    db.commit()
+    return True
diff --git a/backend/db/crud/providers.py b/backend/db/crud/providers.py
new file mode 100644
index 00000000..49388aac
--- /dev/null
+++ b/backend/db/crud/providers.py
@@ -0,0 +1,166 @@
+"""Provider Config CRUD"""
+
+import logging
+from typing import Optional
+
+from sqlalchemy.orm import Session
+
+from db.models import ProviderConfig
+
+logger = logging.getLogger(__name__)
+
+
+def _mask_secret(value: str, visible: int = 4) -> str:
+    """脱敏显示密钥"""
+    if not value:
+        return ""
+    if len(value) <= visible:
+        return "*" * len(value)
+    return "*" * 4 + value[-visible:]
+
+
+def _serialize_provider(p: ProviderConfig) -> dict:
+    """序列化 ProviderConfig 为前端可用 dict（密钥脱敏）"""
+    d = {
+        "id": p.id,
+        "name": p.name,
+        "is_active": p.is_active,
+        "api_key_set": bool(p.api_key),
+        "api_key_hint": _mask_secret(p.api_key),
+        "base_url": p.base_url or "",
+        "model_name": p.model_name or "",
+    }
+    if p.category == "openai":
+        d["light_model_name"] = p.light_model_name or ""
+        d["supports_function_calling"] = p.supports_function_calling
+    elif p.category == "paddleocr":
+        d["use_doc_orientation"] = p.use_doc_orientation
+        d["use_doc_unwarping"] = p.use_doc_unwarping
+        d["use_chart_recognition"] = p.use_chart_recognition
+    return d
+
+
+def get_user_providers(db: Session, user_id: int) -> dict:
+    """获取用户的所有 provider 配置，按类别分组返回"""
+    providers = db.query(ProviderConfig).filter(
+        ProviderConfig.user_id == user_id
+    ).order_by(ProviderConfig.created_at).all()
+
+    result = {
+        "openai_providers": [],
+        "anthropic_providers": [],
+        "paddleocr_providers": [],
+        "active_openai_id": None,
+        "active_anthropic_id": None,
+        "active_paddleocr_id": None,
+    }
+    category_key = {
+        "openai": "openai_providers",
+        "anthropic": "anthropic_providers",
+        "paddleocr": "paddleocr_providers",
+    }
+    active_key = {
+        "openai": "active_openai_id",
+        "anthropic": "active_anthropic_id",
+        "paddleocr": "active_paddleocr_id",
+    }
+
+    for p in providers:
+        key = category_key.get(p.category)
+        if key:
+            result[key].append(_serialize_provider(p))
+        if p.is_active:
+            ak = active_key.get(p.category)
+            if ak:
+                result[ak] = p.id
+
+    return result
+
+
+def save_user_providers(db: Session, user_id: int, data: dict) -> None:
+    """保存用户的 provider 配置（全量覆盖）
+
+    data 结构:
+        openai_providers: [{ id, name, api_key?, base_url, model_name, ... }]
+        anthropic_providers: [...]
+        paddleocr_providers: [...]
+        active_openai_id: str | None
+        active_anthropic_id: str | None
+        active_paddleocr_id: str | None
+    """
+    active_ids = {
+        "openai": data.get("active_openai_id"),
+        "anthropic": data.get("active_anthropic_id"),
+        "paddleocr": data.get("active_paddleocr_id"),
+    }
+
+    # 读取已有配置（用于保留未重新提交的 api_key）
+    existing = {
+        p.id: p for p in db.query(ProviderConfig).filter(
+            ProviderConfig.user_id == user_id
+        ).all()
+    }
+
+    # 收集本次提交的所有 ID
+    submitted_ids = set()
+    items_to_save = []
+
+    category_map = {
+        "openai_providers": "openai",
+        "anthropic_providers": "anthropic",
+        "paddleocr_providers": "paddleocr",
+    }
+
+    for list_key, category in category_map.items():
+        for item in data.get(list_key, []):
+            pid = item.get("id")
+            submitted_ids.add(pid)
+            old = existing.get(pid)
+
+            # API Key：前端提交了新值则更新，否则保留旧值
+            new_api_key = item.get("api_key") or item.get("api_token") or ""
+            if not new_api_key and old:
+                new_api_key = old.api_key or ""
+
+            base_url = item.get("base_url") or item.get("api_url") or ""
+            model_name = item.get("model_name") or item.get("model") or ""
+
+            items_to_save.append({
+                "id": pid,
+                "category": category,
+                "name": item.get("name", ""),
+                "is_active": pid == active_ids.get(category),
+                "api_key": new_api_key,
+                "base_url": base_url,
+                "model_name": model_name,
+                "light_model_name": item.get("light_model_name", ""),
+                "supports_function_calling": item.get("supports_function_calling", True),
+                "use_doc_orientation": item.get("use_doc_orientation", False),
+                "use_doc_unwarping": item.get("use_doc_unwarping", False),
+                "use_chart_recognition": item.get("use_chart_recognition", False),
+            })
+
+    # 删除已移除的 provider
+    for pid in set(existing.keys()) - submitted_ids:
+        db.delete(existing[pid])
+
+    # 更新或新增
+    for item in items_to_save:
+        old = existing.get(item["id"])
+        if old:
+            for k, v in item.items():
+                setattr(old, k, v)
+        else:
+            p = ProviderConfig(user_id=user_id, **item)
+            db.add(p)
+
+    db.commit()
+
+
+def get_active_provider(db: Session, user_id: int, category: str) -> Optional[ProviderConfig]:
+    """获取用户指定类别的激活 provider（用于后端调用 LLM/OCR 时读取凭据）"""
+    return db.query(ProviderConfig).filter(
+        ProviderConfig.user_id == user_id,
+        ProviderConfig.category == category,
+        ProviderConfig.is_active == True,
+    ).first()
diff --git a/backend/db/crud/questions.py b/backend/db/crud/questions.py
new file mode 100644
index 00000000..69c954ae
--- /dev/null
+++ b/backend/db/crud/questions.py
@@ -0,0 +1,477 @@
+"""题目 CRUD"""
+
+import hashlib
+import json
+import logging
+import re
+from datetime import datetime
+from typing import List, Dict, Any, Optional, Tuple
+
+from sqlalchemy.orm import Session, joinedload, selectinload
+
+from db.models import UploadBatch, Question, KnowledgeTag, QuestionTagMapping
+from db.crud.tags import _parse_tag_list, get_or_create_tag
+
+logger = logging.getLogger(__name__)
+
+
+def _get_filters():
+    """延迟导入共享过滤函数，避免循环导入"""
+    from db.crud import _filter_by_subject, _filter_by_user
+    return _filter_by_subject, _filter_by_user
+
+
+def compute_content_hash(content_blocks: List[Dict]) -> str:
+    """
+    基于 content_blocks 计算去重哈希
+    使用题目文本内容计算 SHA256
+    """
+    text_parts = []
+    for block in content_blocks:
+        if block.get("block_type") == "text":
+            text_parts.append(block.get("content", ""))
+
+    text = " ".join(text_parts).strip()
+    if not text:
+        # 如果没有文本内容，使用整个 content_blocks 的 JSON 作为哈希源
+        text = json.dumps(content_blocks, ensure_ascii=False)
+
+    return hashlib.sha256(text.encode()).hexdigest()
+
+
+def question_exists(db, content_hash, user_id=None):
+    """检查题目是否已存在（通过内容哈希 + 用户隔离）"""
+    q = db.query(Question).filter(Question.content_hash == content_hash)
+    if user_id is not None:
+        q = q.filter(Question.user_id == user_id)
+    return q.first()
+
+
+def save_questions_to_db(
+    db: Session,
+    questions: List[Dict],
+    batch_info: Dict[str, Any],
+    user_id=None,
+) -> Dict[str, int]:
+    """
+    批量入库题目
+
+    Args:
+        db: 数据库会话
+        questions: 题目列表（字典格式，来自 questions.json）
+        batch_info: 批次信息，包含 original_filename, file_path 等
+
+    Returns:
+        dict: {"created": 新增数量, "duplicates": 重复数量}
+    """
+    # 科目由编排智能体识别，不再使用关键词匹配
+    subject = batch_info.get("subject") or "未知"
+
+    # 创建批次记录
+    batch = UploadBatch(
+        user_id=user_id,
+        original_filename=batch_info.get("original_filename", "未知"),
+        subject=subject,
+        file_path=batch_info.get("file_path", ""),
+        upload_time=batch_info.get("upload_time") or datetime.utcnow()
+    )
+    db.add(batch)
+    db.flush()  # 获取 batch.id
+
+    created = 0
+    duplicates = 0
+
+    for q in questions:
+        content_blocks = q.get("content_blocks", [])
+        if not content_blocks:
+            continue
+
+        content_hash = compute_content_hash(content_blocks)
+
+        # 检查是否已存在
+        if question_exists(db, content_hash, user_id=user_id):
+            duplicates += 1
+            continue
+
+        # 创建题目记录
+        question = Question(
+            user_id=user_id,
+            batch_id=batch.id,
+            content_hash=content_hash,
+            question_type=q.get("question_type"),
+            content_json=json.dumps(content_blocks, ensure_ascii=False),
+            options_json=json.dumps(q.get("options"), ensure_ascii=False) if q.get("options") else None,
+            has_formula=q.get("has_formula", False),
+            has_image=q.get("has_image", False),
+            image_refs_json=json.dumps(q.get("image_refs"), ensure_ascii=False) if q.get("image_refs") else None,
+            needs_correction=q.get("needs_correction", False),
+            ocr_issues_json=json.dumps(q.get("ocr_issues"), ensure_ascii=False) if q.get("ocr_issues") else None,
+            answer=q.get("answer") or None,
+            user_answer=q.get("user_answer") or None,
+        )
+        db.add(question)
+        db.flush()
+
+        # 处理知识点标签
+        knowledge_tags = q.get("knowledge_tags") or []
+        for tag_name in knowledge_tags:
+            tag = get_or_create_tag(db, tag_name, subject)
+            mapping = QuestionTagMapping(
+                question_id=question.id,
+                tag_id=tag.id
+            )
+            db.add(mapping)
+
+        created += 1
+
+    db.commit()
+
+    return {"created": created, "duplicates": duplicates}
+
+
+def get_questions_by_subject(
+    db: Session,
+    subject: str,
+    limit: int = 100,
+    offset: int = 0,
+    user_id=None,
+) -> List[Question]:
+    """按科目查询题目"""
+    query = db.query(Question).join(UploadBatch).filter(
+        UploadBatch.subject == subject
+    )
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return query.order_by(Question.created_at.desc()).offset(offset).limit(limit).all()
+
+
+def get_questions_by_tag(
+    db: Session,
+    tag_name: str,
+    limit: int = 100,
+    offset: int = 0,
+    user_id=None,
+) -> List[Question]:
+    """按标签查询题目"""
+    query = db.query(Question).join(QuestionTagMapping).join(KnowledgeTag).filter(
+        KnowledgeTag.tag_name == tag_name
+    )
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return query.order_by(Question.created_at.desc()).offset(offset).limit(limit).all()
+
+
+def get_history_questions(
+    db: Session,
+    start_date: Optional[datetime] = None,
+    end_date: Optional[datetime] = None,
+    page: int = 1,
+    page_size: int = 20,
+    user_id=None,
+) -> Tuple[List[Question], int]:
+    """
+    分页查询历史题目（全部题目）
+
+    Args:
+        db: 数据库会话
+        start_date: 开始日期筛选
+        end_date: 结束日期筛选
+        page: 页码（从1开始）
+        page_size: 每页数量
+
+    Returns:
+        (题目列表, 总数)
+    """
+    query = db.query(Question).join(UploadBatch)
+
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+
+    if start_date:
+        query = query.filter(UploadBatch.upload_time >= start_date)
+    if end_date:
+        from datetime import timedelta
+        query = query.filter(UploadBatch.upload_time < end_date + timedelta(days=1))
+
+    # 获取总数
+    total = query.count()
+
+    # 分页查询
+    offset = (page - 1) * page_size
+    questions = (
+        query.options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
+        .order_by(Question.created_at.desc())
+        .offset(offset)
+        .limit(page_size)
+        .all()
+    )
+
+    return questions, total
+
+
+def search_questions(
+    db: Session,
+    keyword: Optional[str] = None,
+    knowledge_tag: Optional[str] = None,
+    question_type: Optional[str] = None,
+    page: int = 1,
+    page_size: int = 20,
+    user_id=None,
+) -> Tuple[List[Question], int]:
+    """
+    搜索题目（知识点/题型/关键字）
+
+    Args:
+        db: 数据库会话
+        keyword: 关键字搜索（匹配题目内容 content_json）
+        knowledge_tag: 知识点标签筛选
+        question_type: 题型筛选
+        page: 页码（从1开始）
+        page_size: 每页数量
+
+    Returns:
+        (题目列表, 总数)
+    """
+    query = db.query(Question)
+
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+
+    # 关键字搜索：匹配 content_json 中的内容
+    if keyword:
+        escaped = re.sub(r"([%_\\])", r"\\\1", keyword)
+        query = query.filter(Question.content_json.ilike(f"%{escaped}%"))
+
+    # 题型筛选
+    if question_type:
+        query = query.filter(Question.question_type == question_type)
+
+    # 知识点标签筛选（支持逗号分隔多选，OR 语义）
+    if knowledge_tag:
+        tag_list = _parse_tag_list(knowledge_tag)
+        if tag_list:
+            query = query.join(QuestionTagMapping).join(KnowledgeTag).filter(
+                KnowledgeTag.tag_name.in_(tag_list)
+            )
+
+    # 获取总数（需要先去除distinct，因为join可能产生重复）
+    total = query.distinct().count()
+
+    # 分页查询
+    offset = (page - 1) * page_size
+    questions = (
+        query.distinct()
+        .options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
+        .order_by(Question.created_at.desc())
+        .offset(offset)
+        .limit(page_size)
+        .all()
+    )
+
+    return questions, total
+
+
+def query_questions(
+    db: Session,
+    subject: Optional[str] = None,
+    knowledge_tag: Optional[str] = None,
+    question_type: Optional[str] = None,
+    keyword: Optional[str] = None,
+    start_date: Optional[datetime] = None,
+    end_date: Optional[datetime] = None,
+    review_status: Optional[str] = None,
+    page: int = 1,
+    page_size: int = 20,
+    user_id=None,
+) -> Tuple[List[Question], int]:
+    """
+    统一查询题目（合并 get_history_questions 和 search_questions 的能力）
+
+    支持所有筛选条件任意组合。
+    """
+    query = db.query(Question).join(UploadBatch)
+
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+
+    # 未筛选的总收录数（仅按用户隔离）
+    grand_total = query.distinct().count()
+
+    if subject:
+        query = query.filter(UploadBatch.subject == subject)
+
+    if question_type:
+        query = query.filter(Question.question_type == question_type)
+
+    if keyword:
+        escaped = re.sub(r"([%_\\])", r"\\\1", keyword)
+        query = query.filter(Question.content_json.ilike(f"%{escaped}%"))
+
+    if knowledge_tag:
+        tag_list = _parse_tag_list(knowledge_tag)
+        if tag_list:
+            query = query.join(QuestionTagMapping).join(KnowledgeTag).filter(
+                KnowledgeTag.tag_name.in_(tag_list)
+            )
+
+    if start_date:
+        query = query.filter(Question.created_at >= start_date)
+    if end_date:
+        from datetime import timedelta
+        query = query.filter(Question.created_at < end_date + timedelta(days=1))
+
+    if review_status:
+        query = query.filter(Question.review_status == review_status)
+
+    total = query.distinct().count()
+
+    offset = (page - 1) * page_size
+    questions = (
+        query.distinct()
+        .options(selectinload(Question.batch), selectinload(Question.tags).selectinload(QuestionTagMapping.tag))
+        .order_by(Question.created_at.desc())
+        .offset(offset)
+        .limit(page_size)
+        .all()
+    )
+
+    return questions, total, grand_total
+
+
+def get_questions_by_ids(db: Session, question_ids: List[int], user_id=None) -> List[Question]:
+    """按 ID 列表批量查询题目"""
+    if not question_ids:
+        return []
+    query = (
+        db.query(Question)
+        .options(joinedload(Question.batch), joinedload(Question.tags).joinedload(QuestionTagMapping.tag))
+        .filter(Question.id.in_(question_ids))
+    )
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return query.all()
+
+
+def delete_question(db: Session, question_id: int, user_id=None) -> bool:
+    """
+    删除题目
+
+    Args:
+        db: 数据库会话
+        question_id: 题目ID
+        user_id: 用户ID（非 None 时校验归属）
+
+    Returns:
+        是否删除成功
+    """
+    query = db.query(Question).filter(Question.id == question_id)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    question = query.first()
+    if not question:
+        return False
+
+    try:
+        # 删除关联的标签映射
+        db.query(QuestionTagMapping).filter(QuestionTagMapping.question_id == question_id).delete()
+
+        # 删除题目
+        db.delete(question)
+        db.commit()
+    except Exception as e:
+        db.rollback()
+        logger.error(f"删除题目 {question_id} 失败: {e}")
+        raise
+
+    return True
+
+
+def update_user_answer(db: Session, question_id: int, user_answer: str, user_id=None) -> Optional[Question]:
+    """更新用户答案"""
+    query = db.query(Question).filter(Question.id == question_id)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    question = query.first()
+    if not question:
+        return None
+
+    try:
+        question.user_answer = user_answer
+        question.updated_at = datetime.utcnow()
+        db.commit()
+        db.refresh(question)
+        return question
+    except Exception as e:
+        db.rollback()
+        logger.error(f"更新题目 {question_id} 答案失败: {e}")
+        raise
+
+
+def update_question_answer(db: Session, question_id: int, answer: str, user_id=None) -> Optional[Question]:
+    """保存/更新题目答案（Markdown 格式）"""
+    query = db.query(Question).filter(Question.id == question_id)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    question = query.first()
+    if not question:
+        return None
+
+    try:
+        question.answer = answer
+        question.updated_at = datetime.utcnow()
+        db.commit()
+        db.refresh(question)
+        return question
+    except Exception as e:
+        db.rollback()
+        logger.error(f"保存题目 {question_id} 答案失败: {e}")
+        raise
+
+
+VALID_REVIEW_STATUSES = ('待复习', '复习中', '已掌握')
+
+
+def update_review_status(db: Session, question_id: int, review_status: str, user_id=None) -> Optional[Question]:
+    """更新题目复习状态"""
+    if review_status not in VALID_REVIEW_STATUSES:
+        raise ValueError(f"无效的复习状态: {review_status}，可选值: {VALID_REVIEW_STATUSES}")
+
+    query = db.query(Question).filter(Question.id == question_id)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    question = query.first()
+    if not question:
+        return None
+
+    try:
+        question.review_status = review_status
+        question.updated_at = datetime.utcnow()
+        db.commit()
+        db.refresh(question)
+        return question
+    except Exception as e:
+        db.rollback()
+        logger.error(f"更新题目 {question_id} 复习状态失败: {e}")
+        raise
+
+
+def get_existing_subjects(db, user_id=None):
+    """获取数据库中已有的所有科目名称（去重）"""
+    query = db.query(UploadBatch.subject).distinct().filter(
+        UploadBatch.subject.isnot(None),
+        UploadBatch.subject != "",
+    )
+    if user_id is not None:
+        query = query.filter(UploadBatch.user_id == user_id)
+    return [r[0] for r in query.all()]
+
+
+def get_existing_question_types(db: Session, user_id=None) -> List[str]:
+    """获取数据库中已有的所有题型（去重）"""
+    query = db.query(Question.question_type).distinct().filter(
+        Question.question_type.isnot(None),
+        Question.question_type != "",
+    )
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return [r[0] for r in query.all()]
diff --git a/backend/db/crud/split_records.py b/backend/db/crud/split_records.py
new file mode 100644
index 00000000..d80e1b4d
--- /dev/null
+++ b/backend/db/crud/split_records.py
@@ -0,0 +1,91 @@
+"""分割历史记录 CRUD"""
+
+import json
+import logging
+from typing import List, Dict, Optional
+
+from sqlalchemy.orm import Session
+from sqlalchemy import func
+
+from db.models import SplitRecord
+
+logger = logging.getLogger(__name__)
+
+
+MAX_SPLIT_RECORDS = 20
+
+
+def save_split_record(
+    db: Session,
+    subject: Optional[str],
+    model_provider: str,
+    file_names: List[str],
+    questions: List[Dict],
+    user_id=None,
+) -> SplitRecord:
+    """保存一次分割操作的完整结果，超过上限自动清理最旧记录"""
+    record = SplitRecord(
+        user_id=user_id,
+        subject=subject,
+        model_provider=model_provider,
+        file_names_json=json.dumps(file_names, ensure_ascii=False),
+        questions_json=json.dumps(questions, ensure_ascii=False),
+        question_count=len(questions),
+    )
+    db.add(record)
+    try:
+        db.commit()
+        db.refresh(record)
+        _cleanup_old_split_records(db, user_id=user_id)
+        return record
+    except Exception as e:
+        db.rollback()
+        logger.error(f"保存分割记录失败: {e}")
+        raise
+
+
+def _cleanup_old_split_records(db: Session, user_id=None):
+    """删除超出上限的最旧分割记录（按用户隔离）"""
+    count_query = db.query(func.count(SplitRecord.id))
+    if user_id is not None:
+        count_query = count_query.filter(SplitRecord.user_id == user_id)
+    count = count_query.scalar()
+    if count <= MAX_SPLIT_RECORDS:
+        return
+    overflow = count - MAX_SPLIT_RECORDS
+    ids_query = db.query(SplitRecord.id)
+    if user_id is not None:
+        ids_query = ids_query.filter(SplitRecord.user_id == user_id)
+    old_ids = [
+        row[0] for row in
+        ids_query.order_by(SplitRecord.created_at.asc())
+        .limit(overflow)
+        .all()
+    ]
+    try:
+        db.query(SplitRecord).filter(SplitRecord.id.in_(old_ids)).delete(synchronize_session=False)
+        db.commit()
+        logger.info(f"已清理 {overflow} 条过期分割记录")
+    except Exception as e:
+        db.rollback()
+        logger.error(f"清理分割记录失败: {e}")
+
+
+def get_recent_split_records(db, limit: int = 10, user_id=None):
+    """获取最近 N 条分割记录（不加载 questions_json 大字段）"""
+    from sqlalchemy.orm import defer
+    query = (
+        db.query(SplitRecord)
+        .options(defer(SplitRecord.questions_json))
+    )
+    if user_id is not None:
+        query = query.filter(SplitRecord.user_id == user_id)
+    return query.order_by(SplitRecord.created_at.desc()).limit(limit).all()
+
+
+def get_split_record_by_id(db: Session, record_id: int, user_id=None) -> Optional[SplitRecord]:
+    """按 ID 获取单条分割记录（含完整 questions_json）"""
+    query = db.query(SplitRecord).filter(SplitRecord.id == record_id)
+    if user_id is not None:
+        query = query.filter(SplitRecord.user_id == user_id)
+    return query.first()
diff --git a/backend/db/crud/stats.py b/backend/db/crud/stats.py
new file mode 100644
index 00000000..375b7734
--- /dev/null
+++ b/backend/db/crud/stats.py
@@ -0,0 +1,272 @@
+"""统计信息 CRUD"""
+
+import logging
+from datetime import datetime
+from typing import List, Dict, Any, Optional
+
+from sqlalchemy.orm import Session
+from sqlalchemy import func
+
+from db.models import UploadBatch, Question, KnowledgeTag, QuestionTagMapping
+from db.crud.questions import VALID_REVIEW_STATUSES
+
+logger = logging.getLogger(__name__)
+
+
+def _get_filters():
+    """延迟导入共享过滤函数，避免循环导入"""
+    from db.crud import _filter_by_subject, _filter_by_user
+    return _filter_by_subject, _filter_by_user
+
+
+def get_statistics(db: Session, subject: Optional[str] = None, user_id=None) -> Dict[str, Any]:
+    """获取统计信息"""
+    _filter_by_subject, _filter_by_user = _get_filters()
+
+    q_query = _filter_by_subject(db.query(func.count(Question.id)), subject)
+    if user_id is not None:
+        q_query = q_query.filter(Question.user_id == user_id)
+    total_questions = q_query.scalar()
+
+    batch_query = db.query(func.count(UploadBatch.id))
+    if user_id is not None:
+        batch_query = batch_query.filter(UploadBatch.user_id == user_id)
+    total_batches = batch_query.scalar()
+
+    total_tags = db.query(func.count(KnowledgeTag.id)).scalar()
+
+    # 按科目统计
+    subject_q = db.query(UploadBatch.subject, func.count(Question.id)).join(Question)
+    if user_id is not None:
+        subject_q = subject_q.filter(Question.user_id == user_id)
+    subject_stats = subject_q.group_by(UploadBatch.subject).all()
+
+    return {
+        "total_questions": total_questions or 0,
+        "total_batches": total_batches or 0,
+        "total_tags": total_tags or 0,
+        "by_subject": {s: c for s, c in subject_stats}
+    }
+
+
+def get_knowledge_stats(db: Session, subject: Optional[str] = None, limit: Optional[int] = None, user_id=None) -> List[Dict]:
+    """
+    获取知识点统计信息
+
+    Args:
+        subject: 可选，按学科筛选
+        limit: 可选，返回前 N 条
+        user_id: 可选，按用户隔离
+
+    Returns:
+        [{"tag_name": "xxx", "count": 10}, ...]
+    """
+    query = db.query(
+        KnowledgeTag.tag_name,
+        func.count(QuestionTagMapping.question_id).label("count")
+    ).join(
+        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
+    )
+
+    if user_id is not None:
+        query = query.join(Question, Question.id == QuestionTagMapping.question_id).filter(
+            Question.user_id == user_id
+        )
+
+    if subject:
+        query = query.filter(KnowledgeTag.subject == subject)
+
+    query = query.group_by(
+        KnowledgeTag.id, KnowledgeTag.tag_name
+    ).order_by(
+        func.count(QuestionTagMapping.question_id).desc()
+    )
+
+    if limit:
+        query = query.limit(limit)
+
+    stats = query.all()
+
+    return [{"tag_name": tag_name, "count": count} for tag_name, count in stats]
+
+
+def get_review_status_stats(db: Session, subject: Optional[str] = None, user_id=None) -> Dict[str, int]:
+    """按复习状态分组统计数量"""
+    _filter_by_subject, _filter_by_user = _get_filters()
+
+    query = _filter_by_subject(db.query(
+        Question.review_status,
+        func.count(Question.id)
+    ), subject)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+
+    rows = query.group_by(Question.review_status).all()
+
+    result = {s: 0 for s in VALID_REVIEW_STATUSES}
+    for status, count in rows:
+        key = status or '待复习'
+        if key in result:
+            result[key] += count
+        else:
+            result['待复习'] += count
+    return result
+
+
+def get_today_mastered_count(db: Session, subject: Optional[str] = None, user_id=None) -> int:
+    """获取今日新掌握的题目数（updated_at 在今天且状态为已掌握）"""
+    _filter_by_subject, _filter_by_user = _get_filters()
+
+    today_start = datetime.utcnow().replace(hour=0, minute=0, second=0, microsecond=0)
+    query = _filter_by_subject(db.query(func.count(Question.id)).filter(
+        Question.updated_at >= today_start,
+        Question.review_status == '已掌握',
+    ), subject)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    return query.scalar() or 0
+
+
+def get_daily_counts(db: Session, days: int = 7, subject: Optional[str] = None, user_id=None) -> List[Dict[str, Any]]:
+    """获取最近 N 天每日新增题目数 + 每日新增已掌握数"""
+    from datetime import timedelta
+    _filter_by_subject, _filter_by_user = _get_filters()
+
+    cutoff = datetime.utcnow().replace(hour=0, minute=0, second=0, microsecond=0) - timedelta(days=days - 1)
+
+    # SQLite 兼容：使用 func.date() 提取日期字符串
+    date_expr = func.date(Question.created_at)
+
+    # 新增题目数
+    query = _filter_by_subject(db.query(
+        date_expr.label('date'),
+        func.count(Question.id).label('count')
+    ).filter(Question.created_at >= cutoff), subject)
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    rows = query.group_by(date_expr).order_by(date_expr).all()
+    date_map = {str(row.date): row.count for row in rows}
+
+    # 每日新增已掌握数（按 updated_at 统计）
+    mastered_date_expr = func.date(Question.updated_at)
+    mq = _filter_by_subject(db.query(
+        mastered_date_expr.label('date'),
+        func.count(Question.id).label('count')
+    ).filter(
+        Question.updated_at >= cutoff,
+        Question.review_status == '已掌握',
+    ), subject)
+    if user_id is not None:
+        mq = mq.filter(Question.user_id == user_id)
+    mastered_rows = mq.group_by(mastered_date_expr).order_by(mastered_date_expr).all()
+    mastered_map = {str(row.date): row.count for row in mastered_rows}
+
+    # 填充缺失的日期
+    result = []
+    for i in range(days):
+        d = cutoff + timedelta(days=i)
+        date_key = d.strftime('%Y-%m-%d')
+        date_str = d.strftime('%m-%d')
+        result.append({
+            'date': date_str,
+            'count': date_map.get(date_key, 0),
+            'mastered': mastered_map.get(date_key, 0),
+        })
+
+    return result
+
+
+def get_tag_status_stats(db: Session, subject: Optional[str] = None, limit: int = 10, user_id=None) -> List[Dict]:
+    """
+    知识点 × 掌握状态统计（堆叠柱状图用）
+
+    Returns:
+        [{"tag_name": "函数", "待复习": 3, "复习中": 2, "已掌握": 5}, ...]
+    """
+    query = db.query(
+        KnowledgeTag.tag_name,
+        Question.review_status,
+        func.count(Question.id).label('count')
+    ).join(
+        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
+    ).join(
+        Question, Question.id == QuestionTagMapping.question_id
+    )
+
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    if subject:
+        query = query.filter(KnowledgeTag.subject == subject)
+
+    rows = query.group_by(
+        KnowledgeTag.tag_name, Question.review_status
+    ).all()
+
+    # 按 tag 聚合
+    tag_data = {}
+    tag_totals = {}
+    for tag_name, status, count in rows:
+        if tag_name not in tag_data:
+            tag_data[tag_name] = {'tag_name': tag_name, **{s: 0 for s in VALID_REVIEW_STATUSES}}
+            tag_totals[tag_name] = 0
+        key = status or '待复习'
+        if key in VALID_REVIEW_STATUSES:
+            tag_data[tag_name][key] += count
+        else:
+            tag_data[tag_name]['待复习'] += count
+        tag_totals[tag_name] += count
+
+    # 按总数降序，取 top N
+    sorted_tags = sorted(tag_data.keys(), key=lambda t: tag_totals[t], reverse=True)[:limit]
+    return [tag_data[t] for t in sorted_tags]
+
+
+def get_tag_type_stats(db: Session, subject: Optional[str] = None, tag_limit: int = 8, user_id=None) -> Dict[str, Any]:
+    """
+    知识点 × 题型交叉统计（热力图用）
+
+    Returns:
+        {"tags": ["函数", ...], "types": ["选择题", ...], "data": [[3, 1, ...], ...]}
+    """
+    query = db.query(
+        KnowledgeTag.tag_name,
+        Question.question_type,
+        func.count(Question.id).label('count')
+    ).join(
+        QuestionTagMapping, QuestionTagMapping.tag_id == KnowledgeTag.id
+    ).join(
+        Question, Question.id == QuestionTagMapping.question_id
+    ).filter(
+        Question.question_type.isnot(None),
+        Question.question_type != '',
+    )
+
+    if user_id is not None:
+        query = query.filter(Question.user_id == user_id)
+    if subject:
+        query = query.filter(KnowledgeTag.subject == subject)
+
+    rows = query.group_by(
+        KnowledgeTag.tag_name, Question.question_type
+    ).all()
+
+    # 收集所有 tag 和 type
+    tag_totals = {}
+    type_set = set()
+    cross = {}
+    for tag_name, q_type, count in rows:
+        tag_totals[tag_name] = tag_totals.get(tag_name, 0) + count
+        type_set.add(q_type)
+        cross[(tag_name, q_type)] = count
+
+    # 按总数取 top N tag
+    sorted_tags = sorted(tag_totals.keys(), key=lambda t: tag_totals[t], reverse=True)[:tag_limit]
+    sorted_types = sorted(type_set)
+
+    # 构建矩阵
+    data = []
+    for tag in sorted_tags:
+        row = [cross.get((tag, t), 0) for t in sorted_types]
+        data.append(row)
+
+    return {"tags": sorted_tags, "types": sorted_types, "data": data}
diff --git a/backend/db/crud/tags.py b/backend/db/crud/tags.py
new file mode 100644
index 00000000..dc16753d
--- /dev/null
+++ b/backend/db/crud/tags.py
@@ -0,0 +1,47 @@
+"""知识点标签 CRUD"""
+
+import logging
+from typing import List, Optional
+
+from sqlalchemy.orm import Session
+
+from db.models import KnowledgeTag
+
+logger = logging.getLogger(__name__)
+
+
+def _parse_tag_list(knowledge_tag: str) -> List[str]:
+    """将逗号分隔的标签字符串拆分为去空白的非空列表"""
+    return [t.strip() for t in knowledge_tag.split(',') if t.strip()]
+
+
+def get_or_create_tag(db: Session, tag_name: str, subject: str) -> KnowledgeTag:
+    """获取或创建知识点标签"""
+    tag = db.query(KnowledgeTag).filter_by(
+        tag_name=tag_name,
+        subject=subject
+    ).first()
+
+    if not tag:
+        tag = KnowledgeTag(tag_name=tag_name, subject=subject)
+        db.add(tag)
+        db.flush()
+
+    return tag
+
+
+def get_existing_tag_names(db: Session, subject: Optional[str] = None) -> List[str]:
+    """获取数据库中已有的知识点标签名称列表（字符串）"""
+    query = db.query(KnowledgeTag.tag_name).distinct()
+    if subject:
+        query = query.filter(KnowledgeTag.subject == subject)
+    rows = query.order_by(KnowledgeTag.tag_name).all()
+    return [r[0] for r in rows]
+
+
+def get_all_tags(db: Session, subject: Optional[str] = None) -> List[KnowledgeTag]:
+    """获取所有标签（可按科目筛选）"""
+    query = db.query(KnowledgeTag)
+    if subject:
+        query = query.filter(KnowledgeTag.subject == subject)
+    return query.order_by(KnowledgeTag.tag_name).all()
diff --git a/backend/db/crud/users.py b/backend/db/crud/users.py
new file mode 100644
index 00000000..5763677b
--- /dev/null
+++ b/backend/db/crud/users.py
@@ -0,0 +1,54 @@
+"""用户认证 CRUD"""
+
+import logging
+from sqlalchemy import func
+
+from db.models import User
+
+logger = logging.getLogger(__name__)
+
+
+def create_user(db, email, password_hash, username, is_admin=False):
+    """创建用户"""
+    user = User(email=email, password_hash=password_hash, username=username, is_admin=is_admin)
+    db.add(user)
+    try:
+        db.commit()
+        db.refresh(user)
+        return user
+    except Exception as e:
+        db.rollback()
+        raise
+
+
+def get_user_by_email(db, email):
+    """按邮箱查询用户"""
+    return db.query(User).filter(User.email == email).first()
+
+
+def get_user_by_id(db, user_id):
+    """按 ID 查询用户"""
+    return db.query(User).filter(User.id == user_id).first()
+
+
+def get_user_by_login(db, identifier):
+    """按邮箱或用户名查询用户（登录用，大小写不敏感邮箱）"""
+    identifier = identifier.strip()
+    return db.query(User).filter(
+        (func.lower(User.email) == identifier.lower()) | (User.username == identifier)
+    ).first()
+
+
+def update_user_password(db, email: str, new_password_hash: str) -> bool:
+    """按邮箱更新用户密码，成功返回 True"""
+    user = db.query(User).filter(User.email == email).first()
+    if not user:
+        return False
+    user.password_hash = new_password_hash
+    user.session_version = (user.session_version or 0) + 1
+    try:
+        db.commit()
+        return True
+    except Exception:
+        db.rollback()
+        raise
diff --git a/backend/db/migrate.py b/backend/db/migrate.py
index f613ea8c..80377d5f 100644
--- a/backend/db/migrate.py
+++ b/backend/db/migrate.py
@@ -12,16 +12,32 @@
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
 from werkzeug.security import generate_password_hash
-from config import settings
+from core.config import settings
 from db import engine, SessionLocal
 from db.models import Base
 from db import crud
 
 
+def _add_column_if_missing(conn, table: str, column: str, col_type: str, default=None):
+    """安全地给已有表添加列，若已存在则跳过"""
+    from sqlalchemy import text, inspect
+    inspector = inspect(conn)
+    existing = [c["name"] for c in inspector.get_columns(table)]
+    if column not in existing:
+        default_clause = f" DEFAULT {default}" if default is not None else ""
+        conn.execute(text(f"ALTER TABLE {table} ADD COLUMN {column} {col_type}{default_clause}"))
+        print(f"[migrate] 已添加列: {table}.{column}")
+
+
 def migrate():
     """增量迁移：仅创建缺失的表，并确保 Admin 用户存在。每次启动自动调用，安全幂等。"""
     Base.metadata.create_all(bind=engine)  # checkfirst=True by default，已存在的表不动
 
+    # 增量列迁移（新增字段时在此追加）
+    with engine.connect() as conn:
+        _add_column_if_missing(conn, "users", "session_version", "INTEGER", 0)
+        conn.commit()
+
     default_users = [
         {"email": "admin@admin.com",  "username": "Admin",  "password": "123456", "is_admin": True},
         {"email": "admin2@admin.com", "username": "Admin2", "password": "123456", "is_admin": True},
diff --git a/backend/db/models.py b/backend/db/models.py
index d9ace4ad..be8d6638 100644
--- a/backend/db/models.py
+++ b/backend/db/models.py
@@ -19,12 +19,14 @@ class User(Base):
     email = Column(String(255), unique=True, nullable=False, index=True)
     password_hash = Column(String(255), nullable=False)
     is_admin = Column(Boolean, default=False)
+    session_version = Column(Integer, default=0, nullable=False)
     created_at = Column(DateTime, default=datetime.utcnow)
 
     questions = relationship("Question", back_populates="user")
     upload_batches = relationship("UploadBatch", back_populates="user")
     split_records = relationship("SplitRecord", back_populates="user")
     provider_configs = relationship("ProviderConfig", back_populates="user", cascade="all, delete-orphan")
+    notes = relationship("Note", back_populates="user")
 
 
 class ProviderConfig(Base):
@@ -100,14 +102,17 @@ class Question(Base):
 
 
 class ChatSession(Base):
-    """教学辅导对话会话"""
+    """对话会话（可绑定题目，也可独立对话）"""
     __tablename__ = "chat_sessions"
 
     id = Column(Integer, primary_key=True)
-    question_id = Column(Integer, ForeignKey("questions.id"), nullable=False, index=True)
+    user_id = Column(Integer, ForeignKey("users.id"), nullable=True, index=True)
+    question_id = Column(Integer, ForeignKey("questions.id"), nullable=True, index=True)
+    title = Column(String(255), default="新对话")
     created_at = Column(DateTime, default=datetime.utcnow)
     updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
 
+    user = relationship("User")
     question = relationship("Question", back_populates="chat_sessions")
     messages = relationship("ChatMessage", back_populates="session", order_by="ChatMessage.id")
 
@@ -164,3 +169,45 @@ class QuestionTagMapping(Base):
 
     question = relationship("Question", back_populates="tags")
     tag = relationship("KnowledgeTag")
+
+
+class Note(Base):
+    """笔记表"""
+    __tablename__ = "notes"
+
+    id = Column(Integer, primary_key=True)
+    user_id = Column(Integer, ForeignKey("users.id"), nullable=True, index=True)
+    title = Column(String(255), nullable=False)          # 笔记标题
+    subject = Column(String(50))                          # 科目
+    content_markdown = Column(Text, default="")           # LLM 整理后的 Markdown 内容
+    source_images_json = Column(Text)                     # 原始上传图片路径列表 JSON
+    ocr_text = Column(Text)                               # OCR 识别的原始文本（保留用于重新整理）
+    created_at = Column(DateTime, default=datetime.utcnow, index=True)
+    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
+
+    user = relationship("User", back_populates="notes")
+    tags = relationship("NoteTagMapping", back_populates="note", cascade="all, delete-orphan")
+
+
+class NoteTagMapping(Base):
+    """笔记-标签关联表（与错题共享 KnowledgeTag）"""
+    __tablename__ = "note_tag_mapping"
+
+    note_id = Column(Integer, ForeignKey("notes.id"), primary_key=True)
+    tag_id = Column(Integer, ForeignKey("knowledge_tags.id"), primary_key=True)
+
+    note = relationship("Note", back_populates="tags")
+    tag = relationship("KnowledgeTag")
+
+
+class EmailVerification(Base):
+    """注册邮箱验证码（仅存哈希，不存明文）"""
+    __tablename__ = "email_verifications"
+
+    id = Column(Integer, primary_key=True)
+    email = Column(String(255), unique=True, nullable=False, index=True)
+    code_hash = Column(String(64), nullable=False)
+    expires_at = Column(DateTime, nullable=False)
+    last_sent_at = Column(DateTime, nullable=True)
+    attempts = Column(Integer, default=0)
+    created_at = Column(DateTime, default=datetime.utcnow)
diff --git a/backend/models/blocks.py b/backend/models/blocks.py
new file mode 100644
index 00000000..1684afcb
--- /dev/null
+++ b/backend/models/blocks.py
@@ -0,0 +1,135 @@
+"""
+基础网络模块：CBAM 注意力、编码/解码块，供 Generator 和 Discriminator 组合使用。
+"""
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class CBAM(nn.Module):
+    """Convolutional Block Attention Module：通道注意力 + 空间注意力。
+
+    Args:
+        in_channels: 输入特征图通道数
+        reduction:   通道注意力 MLP 的压缩比，越大参数越少但表达力越弱
+    """
+    def __init__(self, in_channels: int, reduction: int = 16):
+        super().__init__()
+        self.avg_pool = nn.AdaptiveAvgPool2d(1)
+        self.max_pool = nn.AdaptiveMaxPool2d(1)
+        self.mlp = nn.Sequential(
+            nn.Linear(in_channels, in_channels // reduction, bias=False),
+            nn.ReLU(inplace=True),
+            nn.Linear(in_channels // reduction, in_channels, bias=False),
+        )
+        self.sigmoid = nn.Sigmoid()
+        # 空间注意力：7×7 大感受野，无 BN（标准 CBAM）
+        self.spatial = nn.Sequential(
+            nn.Conv2d(2, 1, kernel_size=7, padding=3, bias=False),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        b, c, h, w = x.shape
+
+        # 通道注意力
+        avg_out = self.mlp(self.avg_pool(x).view(b, c)).view(b, c, 1, 1)
+        max_out = self.mlp(self.max_pool(x).view(b, c)).view(b, c, 1, 1)
+        x = x * self.sigmoid(avg_out + max_out)
+
+        # 空间注意力
+        avg_s = torch.mean(x, dim=1, keepdim=True)
+        max_s, _ = torch.max(x, dim=1, keepdim=True)
+        x = x * self.spatial(torch.cat([avg_s, max_s], dim=1))
+
+        return x
+
+
+class DilatedConvBlock(nn.Module):
+    """空洞卷积块，用于 RefineNet 解码器，扩大感受野而不损失分辨率。
+
+    Args:
+        dilation: 空洞率，默认 2；padding 自动计算以保持尺寸不变
+    """
+    def __init__(self, in_channels: int, out_channels: int, dilation: int = 2):
+        super().__init__()
+        padding = dilation * (3 - 1) // 2
+        self.block = nn.Sequential(
+            nn.Conv2d(in_channels, out_channels, 3,
+                      padding=padding, dilation=dilation, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(inplace=True),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.block(x)
+
+
+class DownSample(nn.Module):
+    """U-Net 编码器下采样块：步长为 2 的卷积，分辨率减半。"""
+    def __init__(self, in_channels: int, out_channels: int):
+        super().__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(in_channels, out_channels, 4, stride=2, padding=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.LeakyReLU(0.2, inplace=True),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.conv(x)
+
+
+class ResBlock(nn.Module):
+    """残差块，参照 EraseNet Residual：conv1(stride) → ReLU → conv2 + skip → BN → ReLU。
+
+    stride=1 时要求 in_channels == out_channels（无投影）；
+    stride=2 时自动添加 1×1 skip 投影以匹配通道和尺寸。
+    """
+    def __init__(self, in_channels: int, out_channels: int, stride: int = 1):
+        super().__init__()
+        self.conv1 = nn.Conv2d(in_channels, in_channels, 3, stride=stride, padding=1)
+        self.conv2 = nn.Conv2d(in_channels, out_channels, 3, padding=1)
+        self.bn    = nn.BatchNorm2d(out_channels)
+        self.skip  = (nn.Conv2d(in_channels, out_channels, 1, stride=stride)
+                      if (stride != 1 or in_channels != out_channels) else None)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        out = F.relu(self.conv1(x))
+        out = self.conv2(out)
+        identity = self.skip(x) if self.skip is not None else x
+        return F.relu(self.bn(out + identity))
+
+
+class LateralConnection(nn.Module):
+    """跳跃连接特征精炼：1×1 → 3×3 → 3×3 → 1×1，参照 EraseNet lateral_connection。
+
+    输入输出通道数相同，中间扩张为 2×channels。
+    """
+    def __init__(self, channels: int):
+        super().__init__()
+        inner = channels * 2
+        self.net = nn.Sequential(
+            nn.Conv2d(channels, channels, 1),
+            nn.Conv2d(channels, inner,    3, padding=1),
+            nn.Conv2d(inner,    inner,    3, padding=1),
+            nn.Conv2d(inner,    channels, 1),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.net(x)
+
+
+class UpSample(nn.Module):
+    """U-Net 解码器上采样块：转置卷积，分辨率翻倍，可选 CBAM 注意力。"""
+    def __init__(self, in_channels: int, out_channels: int,
+                 use_cbam: bool = False, reduction: int = 16):
+        super().__init__()
+        self.conv = nn.Sequential(
+            nn.ConvTranspose2d(in_channels, out_channels, 4, stride=2, padding=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(inplace=True),
+        )
+        self.cbam = CBAM(out_channels, reduction) if use_cbam else nn.Identity()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.cbam(self.conv(x))
diff --git a/backend/models/config.py b/backend/models/config.py
deleted file mode 100644
index 07cd972d..00000000
--- a/backend/models/config.py
+++ /dev/null
@@ -1,59 +0,0 @@
-import os
-import yaml
-from pathlib import Path
-
-
-class Config:
-    _instance = None
-    _config = None
-
-    def __new__(cls):
-        if cls._instance is None:
-            cls._instance = super().__new__(cls)
-        return cls._instance
-
-    def __init__(self):
-        if self._config is None:
-            self._load_config()
-
-    def _load_config(self, config_path=None):
-        if config_path is None:
-            config_path = os.path.join(
-                os.path.dirname(__file__),
-                'config.yaml'
-            )
-
-        if not os.path.exists(config_path):
-            raise FileNotFoundError(f"配置文件不存在: {config_path}")
-
-        with open(config_path, 'r', encoding='utf-8') as f:
-            self._config = yaml.safe_load(f)
-
-    @property
-    def train(self):
-        return self._config.get('train', {})
-
-    @property
-    def loss(self):
-        return self._config.get('loss', {})
-
-    @property
-    def inference(self):
-        return self._config.get('inference', {})
-
-    @property
-    def wandb(self):
-        return self._config.get('wandb', {})
-
-    def get(self, key, default=None):
-        keys = key.split('.')
-        value = self._config
-        for k in keys:
-            if isinstance(value, dict):
-                value = value.get(k, default)
-            else:
-                return default
-        return value
-
-
-config = Config()
diff --git a/backend/models/config.yaml b/backend/models/config.yaml
deleted file mode 100644
index 164f9193..00000000
--- a/backend/models/config.yaml
+++ /dev/null
@@ -1,51 +0,0 @@
-# EnsExam 配置文件
-
-# ====================== 训练参数 ======================
-train:
-  epochs: 100           # 训练轮数
-  batch_size: 4         # 批次大小
-  lr: 0.0001            # 学习率
-  device: "cuda"        # 训练设备 ("cuda" 或 "cpu")
-  num_workers: 4       # 数据加载线程数
-
-  # 路径配置
-  data_root: "D:\\PythonProject1\\LLM\\adcj\\src\\adcj\\EnsExam\\SCUT-EnsExam\\SCUT-EnsExam"
-  save_dir: "./ensexam_checkpoints"
-  resume_path: "./ensexam_checkpoints/latest.pth"
-
-# ====================== 损失权重 ======================
-loss:
-  # LR Loss 多尺度权重 [Ic4, Ic2, Ic1, Ire]
-  lambda_n: [5, 6, 8, 10]
-  beta_n: [0.8, 0.8, 0.8, 2]
-
-  # 各损失总权重
-  lambda_lr: 1.0       # LR损失权重
-  lambda_p: 0.05       # 感知损失权重
-  lambda_style: 120    # 风格损失权重
-  lambda_sn: 1         # SN损失权重
-  lambda_b: 0.4        # 块损失权重
-
-# ====================== 推理参数 ======================
-inference:
-  model_path: "D:\\PythonProject1\\LLM\\adcj\\src\\adcj\\EnsExam\\ensexam_checkpoints\\ensexam_epoch_5.pth"
-  img_size: 512         # 必须和训练时一致
-
-  # 单张图片推理
-  input_image_path: "D:\\PythonProject1\\LLM\\adcj\\src\\adcj\\EnsExam\\SCUT-EnsExam\\SCUT-EnsExam\\test\\all_images\\15.jpg"
-  output_image_path: "./test_output.jpg"
-
-  # 批量目录推理
-  input_dir: "D:\\PythonProject1\\LLM\\adcj\\src\\adcj\\EnsExam\\SCUT-EnsExam\\SCUT-EnsExam\\test\\all_images"
-  output_dir: "./output"
-  patch_size: 512       # 滑动窗口尺寸
-  overlap: 32           # 重叠像素
-
-# ====================== WandB 配置 ======================
-wandb:
-  enabled: true         # 是否启用 wandb
-  project: "ensexam"    # 项目名称
-  name: null           # 实验名称 (null 则自动生成)
-  entity: null         # 团队/用户名
-  offline: true        # 离线模式 (true 使用 offline wandb)
-  dir: "./wandb"       # wandb 日志目录
diff --git a/backend/models/inference.py b/backend/models/inference.py
index e06c848e..3ad48d14 100644
--- a/backend/models/inference.py
+++ b/backend/models/inference.py
@@ -12,8 +12,14 @@
 import threading
 
 import numpy as np
+import torch
 from PIL import Image
 
+from models.model import Generator
+
+# cuDNN v8 frontend 与 A30+PyTorch2.1 不兼容
+torch.backends.cudnn.enabled = False
+
 logger = logging.getLogger(__name__)
 
 _PATCH_SIZE = 512
@@ -50,23 +56,18 @@ def _load_model(self):
             if self._model is not None:
                 return
 
-            import torch
-            torch.backends.cudnn.deterministic = True
-            torch.backends.cudnn.benchmark = False
-            from models.model import Generator
-            from config import settings
+            from core.config import settings
 
             model_path = settings.model_path
             if not model_path.exists():
                 raise FileNotFoundError(
-                    f"模型权重不存在，请将 latest.pth 放到: {model_path}"
+                    f"模型权重不存在，请将 best.pth 放到: {model_path}"
                 )
 
             device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
             generator = Generator().to(device)
 
             ckpt = torch.load(model_path, map_location=device, weights_only=False)
-            # train.py 保存格式：{'G_state_dict': ..., 'D_state_dict': ..., ...}
             state_dict = (
                 ckpt.get("G_state_dict")
                 or ckpt.get("generator_state_dict")
@@ -85,7 +86,7 @@ def _load_model(self):
 
     @staticmethod
     def _preprocess(pil_img: Image.Image):
-        """PIL → float Tensor [-1, 1]，并 padding 到 512 的整数倍
+        """PIL → float ndarray [-1, 1]，并 padding 到 512 的整数倍
 
         Returns:
             arr_padded (np.ndarray): shape (H_pad, W_pad, 3)，float32 in [-1,1]
@@ -105,14 +106,12 @@ def _preprocess(pil_img: Image.Image):
         arr = np.array(img, dtype=np.float32) / 127.5 - 1.0
         return arr, orig_w, orig_h
 
-    @staticmethod
-    def _to_tensor(arr: np.ndarray, device):
+    def _to_tensor(self, arr: np.ndarray) -> torch.Tensor:
         """(H, W, 3) ndarray → (1, 3, H, W) Tensor on device"""
-        import torch
-        return torch.from_numpy(arr).permute(2, 0, 1).unsqueeze(0).to(device)
+        return torch.from_numpy(arr).permute(2, 0, 1).unsqueeze(0).to(self._device)
 
     @staticmethod
-    def _to_numpy(tensor) -> np.ndarray:
+    def _to_numpy(tensor: torch.Tensor) -> np.ndarray:
         """(1, 3, H, W) Tensor → (H, W, 3) uint8 ndarray"""
         arr = tensor.squeeze(0).permute(1, 2, 0).cpu().numpy()
         return np.clip((arr + 1.0) * 127.5, 0, 255).astype(np.uint8)
@@ -123,8 +122,6 @@ def _to_numpy(tensor) -> np.ndarray:
 
     def _infer_sliding_window(self, arr: np.ndarray) -> np.ndarray:
         """滑动窗口推理，patch 重叠区域做加权平均以消除接缝"""
-        import torch
-
         h, w, _ = arr.shape
         result = np.zeros((h, w, 3), dtype=np.float32)
         weight = np.zeros((h, w, 1), dtype=np.float32)
@@ -141,9 +138,8 @@ def _ticks(total):
             for y in _ticks(h):
                 for x in _ticks(w):
                     patch = arr[y:y + _PATCH_SIZE, x:x + _PATCH_SIZE]
-                    tensor = self._to_tensor(patch, self._device)
+                    tensor = self._to_tensor(patch)
                     *_, Icomp = self._model(tensor)
-                    # 直接从 tensor 提取 float32，避免 uint8 截断精度损失
                     out = (Icomp.squeeze(0).permute(1, 2, 0).cpu().numpy() + 1.0) * 127.5
                     result[y:y + _PATCH_SIZE, x:x + _PATCH_SIZE] += out
                     weight[y:y + _PATCH_SIZE, x:x + _PATCH_SIZE] += 1.0
@@ -167,7 +163,6 @@ def run(self, image_bytes: bytes) -> Image.Image:
 
         pil_img = Image.open(io.BytesIO(image_bytes))
         arr, orig_w, orig_h = self._preprocess(pil_img)
-        h, w, _ = arr.shape
 
         result_arr = self._infer_sliding_window(arr)
 
diff --git a/backend/models/model.py b/backend/models/model.py
index 436f19ad..cc931c52 100644
--- a/backend/models/model.py
+++ b/backend/models/model.py
@@ -1,473 +1,213 @@
-import math
-
-import cv2
-import numpy as np
+"""
+生成器网络：CoarseNet（粗擦除 + 掩码预测） + RefineNet（精细修复）。
+
+结构对照 EraseNet (STRnet2)，核心改动：
+  - 编码器：EraseNet 残差块风格（conv1/conva/convb + ResBlock×8）
+  - 跳跃连接：LateralConnection 特征精炼（替代简单 cat）
+  - 掩码分支：从 x_mask（瓶颈前 H/16 特征）出发，与 EraseNet 对齐
+  - 保留：CBAM 注意力、双 sigmoid 掩码（Ms/Mb）、RefineNet 输入含 Ms
+  - RefineNet：EraseNet 风格多尺度空洞卷积，输入保持 cat([Iin, Ms, Ic1])
+"""
 import torch
+import torch.nn as nn
 import torch.nn.functional as F
-from torch import nn
-from torchvision.models import vgg16
-
-
-class CBAM(nn.Module):
-    def __init__(self, in_channels, reduction=16):
-        super().__init__()
-        # 通道注意力（标准实现）
-        self.avg_pool = nn.AdaptiveAvgPool2d(1)
-        self.max_pool = nn.AdaptiveMaxPool2d(1)
-        self.mlp = nn.Sequential(
-            nn.Linear(in_channels, in_channels // reduction, bias=False),
-            nn.ReLU(inplace=True),
-            nn.Linear(in_channels // reduction, in_channels, bias=False)
-        )
-        self.sigmoid = nn.Sigmoid()
-
-        # 空间注意力（关键修正：7×7卷积 + 无BatchNorm）
-        self.spatial = nn.Sequential(
-            nn.Conv2d(2, 1, kernel_size=7, padding=3, bias=False),  # ✅ 7×7
-            nn.Sigmoid()  # ✅ 无BatchNorm
-        )
-
-    def forward(self, x):
-        b, c, h, w = x.shape
-
-        # ========== 通道注意力 ==========
-        avg_out = self.mlp(self.avg_pool(x).view(b, c)).view(b, c, 1, 1)
-        max_out = self.mlp(self.max_pool(x).view(b, c)).view(b, c, 1, 1)
-        channel_att = self.sigmoid(avg_out + max_out)
-        x = x * channel_att  # 广播相乘 [B,C,H,W] * [B,C,1,1]
 
-        # ========== 空间注意力 ==========
-        avg_out = torch.mean(x, dim=1, keepdim=True)  # [B,1,H,W]
-        max_out, _ = torch.max(x, dim=1, keepdim=True)  # [B,1,H,W]
-        spatial_att = self.spatial(torch.cat([avg_out, max_out], dim=1))  # [B,1,H,W]
-        x = x * spatial_att  # 广播相乘 [B,C,H,W] * [B,1,H,W]
+from models.blocks import (DownSample, UpSample, DilatedConvBlock,
+                            ResBlock, LateralConnection)
 
-        return x
-
-
-# 空洞卷积块（Refine网络用）
-class DilatedConvBlock(nn.Module):
-    def __init__(self, in_channels, out_channels, dilation=2):
-        super().__init__()
-        padding = dilation * (3 - 1) // 2
-        self.block = nn.Sequential(
-            nn.Conv2d(in_channels, out_channels, 3, padding=padding, dilation=dilation, bias=False),
-            nn.BatchNorm2d(out_channels),
-            nn.ReLU(inplace=True)
-        )
-
-    def forward(self, x):
-        return self.block(x)
-
-
-# U-Net下采样块（Coarse/Refine网络编码器）
-class DownSample(nn.Module):
-    def __init__(self, in_channels, out_channels):
-        super().__init__()
-        self.conv = nn.Sequential(
-            nn.Conv2d(in_channels, out_channels, 4, 2, 1, bias=False),
-            nn.BatchNorm2d(out_channels),
-            nn.LeakyReLU(0.2, inplace=True)
-        )
-
-    def forward(self, x):
-        x = self.conv(x)
-        return x
-
-
-# U-Net上采样块（Coarse网络解码器）
-class UpSample(nn.Module):
-    def __init__(self, in_channels, out_channels, use_cbam=False):
-        super().__init__()
-        self.conv = nn.Sequential(
-            nn.ConvTranspose2d(in_channels, out_channels, 4, 2, 1, bias=False),
-            nn.BatchNorm2d(out_channels),
-            nn.ReLU(inplace=True)
-        )
-        self.cbam = CBAM(out_channels) if use_cbam else nn.modules.Identity()
 
-    def forward(self, x):
-        x = self.conv(x)
-        x = self.cbam(x)
-        return x
-
-
-# 带空洞卷积的上采样块（Refine网络解码器）
-class DilatedUpSample(nn.Module):
-    def __init__(self, in_channels, out_channels, dilation=2):
-        super().__init__()
-        self.conv = nn.Sequential(
-            nn.ConvTranspose2d(in_channels, out_channels, 4, 2, 1, bias=False),
-            nn.BatchNorm2d(out_channels),
-            nn.ReLU(inplace=True)
-        )
-        self.dilated = DilatedConvBlock(out_channels, out_channels, dilation)
-
-    def forward(self, x):
-        x = self.conv(x)
-        x = self.dilated(x)
-        return x
-
-
-# Coarse网络
 class CoarseNet(nn.Module):
-    def __init__(self, in_channels=3):
-        super().__init__()
-        # 编码器：下采样+CBAM
-        self.down1 = DownSample(in_channels, 64)  # 1/2
-        self.down2 = DownSample(64, 128)  # 1/4
-        self.down3 = DownSample(128, 256)  # 1/8
-        self.down4 = DownSample(256, 512)  # 1/16
-        self.down5 = DownSample(512, 512)  # 1/32
-        # 解码器：上采样
-        self.up1 = UpSample(512, 512, use_cbam=True)  # 1/16
-        self.up2 = UpSample(1024, 256, use_cbam=True)  # 1/8
-        self.up3 = UpSample(512, 128, use_cbam=True)  # 1/4 → Ic4
-        self.up4 = UpSample(256, 64, use_cbam=True)  # 1/2 → Ic2
-        self.up5 = UpSample(128, 64, use_cbam=True)  # 1/1 → Ic1
-        # 输出层：Ms(笔画掩码), Mb(文本块掩码), 多尺度Ic
-        self.up1_seg = UpSample(512, 512, use_cbam=True)
-        self.up2_seg = UpSample(1024, 256, use_cbam=True)
-        self.up3_seg = UpSample(512, 128, use_cbam=True)
-        self.up4_seg = UpSample(256, 64, use_cbam=True)
-        self.up5_seg = UpSample(128, 64, use_cbam=True)
-
-        self.out_ms = nn.Conv2d(64, 1, 3, 1, 1, bias=True)
-        self.out_mb = nn.Conv2d(64, 1, 3, 1, 1, bias=True)
-        self.out_ic4 = nn.Conv2d(128, 3, 3, 1, 1, bias=False)
-        self.out_ic2 = nn.Conv2d(64, 3, 3, 1, 1, bias=False)
-        self.out_ic1 = nn.Conv2d(64, 3, 3, 1, 1, bias=False)
-
-    def forward(self, x):
-        # 编码器（下采样）
-        d1 = self.down1(x)  # H/2
-        d2 = self.down2(d1)  # H/4
-        d3 = self.down3(d2)  # H/8
-        d4 = self.down4(d3)  # H/16
-        d5 = self.down5(d4)  # H/32
-
-        # 解码器（上采样 + 跳跃连接）
-        u1 = self.up1(d5)
-        u1 = torch.cat([u1, d4], dim=1)  # 拼接H/16特征
+    """粗擦除网络：EraseNet 残差编码器 + 双解码器（修复 + 掩码）。
 
-        u2 = self.up2(u1)
-        u2 = torch.cat([u2, d3], dim=1)  # 拼接H/8特征
-
-        u3 = self.up3(u2)  # H/4 → 输出 Ic4
-        Ic4 = torch.tanh(self.out_ic4(u3))
-        u3 = torch.cat([u3, d2], dim=1)  # 拼接H/4特征
-
-        u4 = self.up4(u3)  # H/2 → 输出 Ic2
-        Ic2 = torch.tanh(self.out_ic2(u4))
-        u4 = torch.cat([u4, d1], dim=1)  # 拼接H/2特征
-
-        u5 = self.up5(u4)  # H → 输出 Ic1, Mb, Ms
-        Ic1 = torch.tanh(self.out_ic1(u5))
-
-        u1_seg = self.up1_seg(d5)
-        u1_seg = torch.cat([u1_seg, d4], dim=1)
-        u2_seg = self.up2_seg(u1_seg)
-        u2_seg = torch.cat([u2_seg, d3], dim=1)
-        u3_seg = self.up3_seg(u2_seg)
-        u3_seg = torch.cat([u3_seg, d2], dim=1)
-        u4_seg = self.up4_seg(u3_seg)
-        u4_seg = torch.cat([u4_seg, d1], dim=1)
-        u5_seg = self.up5_seg(u4_seg)
-        Mb = torch.sigmoid(self.out_mb(u5_seg))
-        Ms = torch.sigmoid(self.out_ms(u5_seg))
+    Args:
+        in_channels:    输入通道数，RGB 固定为 3
+        cbam_reduction: CBAM 通道压缩比
+    """
+    def __init__(self, in_channels: int = 3, cbam_reduction: int = 16):
+        super().__init__()
+        r = cbam_reduction
+
+        # ── 编码器（EraseNet 风格） ──────────────────────────────────────
+        self.conv1 = nn.Sequential(
+            nn.Conv2d(in_channels, 32, 4, stride=2, padding=1, bias=False),
+            nn.BatchNorm2d(32), nn.LeakyReLU(0.2, inplace=True))   # H/2,  32
+        self.conva = nn.Sequential(
+            nn.Conv2d(32, 32, 3, padding=1, bias=False),
+            nn.BatchNorm2d(32), nn.LeakyReLU(0.2, inplace=True))   # H/2,  32
+        self.convb = nn.Sequential(
+            nn.Conv2d(32, 64, 4, stride=2, padding=1, bias=False),
+            nn.BatchNorm2d(64), nn.LeakyReLU(0.2, inplace=True))   # H/4,  64
+        self.res1  = ResBlock(64,  64)
+        self.res2  = ResBlock(64,  64)
+        self.res3  = ResBlock(64,  128, stride=2)                   # H/8,  128
+        self.res4  = ResBlock(128, 128)
+        self.res5  = ResBlock(128, 256, stride=2)                   # H/16, 256
+        self.res6  = ResBlock(256, 256)
+        self.res7  = ResBlock(256, 512, stride=2)                   # H/32, 512
+        self.res8  = ResBlock(512, 512)
+        self.conv2 = nn.Conv2d(512, 512, 1)                         # H/32, 512（瓶颈）
+
+        # ── 修复解码器（LateralConnection + CBAM UpSample） ─────────────
+        self.lat1    = LateralConnection(256)
+        self.lat2    = LateralConnection(128)
+        self.lat3    = LateralConnection(64)
+        self.lat4    = LateralConnection(32)
+        self.up1     = UpSample(512, 256, use_cbam=True, reduction=r)
+        self.up2     = UpSample(512, 128, use_cbam=True, reduction=r)
+        self.up3     = UpSample(256, 64,  use_cbam=True, reduction=r)
+        self.up4     = UpSample(128, 32,  use_cbam=True, reduction=r)
+        self.up5     = UpSample(64,  32,  use_cbam=True, reduction=r)
+        self.out_ic4 = nn.Conv2d(64, 3, 1)
+        self.out_ic2 = nn.Conv2d(32, 3, 1)
+        self.out_ic1 = nn.Conv2d(32, 3, 3, padding=1)
+
+        # ── 掩码解码器（从瓶颈 d5 出发，保留全局上下文，双 sigmoid 头） ──
+        self.mask_up_0   = UpSample(512, 256, use_cbam=True, reduction=r)  # H/32→H/16
+        self.mask_up_a   = UpSample(512, 256, use_cbam=True, reduction=r)
+        self.mask_conv_a = nn.Sequential(
+            nn.Conv2d(256, 128, 3, padding=1, bias=False),
+            nn.BatchNorm2d(128), nn.ReLU(inplace=True))
+        self.mask_up_b   = UpSample(256, 128, use_cbam=True, reduction=r)
+        self.mask_conv_b = nn.Sequential(
+            nn.Conv2d(128, 64, 3, padding=1, bias=False),
+            nn.BatchNorm2d(64), nn.ReLU(inplace=True))
+        self.mask_up_c   = UpSample(128, 64, use_cbam=True, reduction=r)
+        self.mask_conv_c = nn.Sequential(
+            nn.Conv2d(64, 32, 3, padding=1, bias=False),
+            nn.BatchNorm2d(32), nn.ReLU(inplace=True))
+        self.mask_up_d   = UpSample(64, 32, use_cbam=True, reduction=r)
+        self.out_ms      = nn.Conv2d(32, 1, 1)                      # 软笔画掩码
+        self.out_mb      = nn.Conv2d(32, 1, 1)                      # 文本块掩码
+
+    def forward(self, x: torch.Tensor):
+        # 编码
+        x      = self.conv1(x)
+        x      = self.conva(x)
+        con_x1 = x                                   # H/2,  32
+        x      = self.convb(x)
+        x      = self.res1(x)
+        con_x2 = x                                   # H/4,  64
+        x      = self.res2(x)
+        x      = self.res3(x)
+        con_x3 = x                                   # H/8,  128
+        x      = self.res4(x)
+        x      = self.res5(x)
+        con_x4 = x                                   # H/16, 256
+        x      = self.res6(x)
+        x      = self.res7(x)
+        x      = self.res8(x)
+        d5     = self.conv2(x)                       # H/32, 512（瓶颈）
+
+        # 修复解码
+        u1  = torch.cat([self.lat1(con_x4), self.up1(d5)],  dim=1)  # H/16, 512
+        u2  = torch.cat([self.lat2(con_x3), self.up2(u1)],  dim=1)  # H/8,  256
+        xo1 = self.up3(u2)                                           # H/4,  64
+        Ic4 = torch.tanh(self.out_ic4(xo1))
+        u3  = torch.cat([self.lat3(con_x2), xo1], dim=1)            # H/4,  128
+        xo2 = self.up4(u3)                                           # H/2,  32
+        Ic2 = torch.tanh(self.out_ic2(xo2))
+        u4  = torch.cat([self.lat4(con_x1), xo2], dim=1)            # H/2,  64
+        Ic1 = torch.tanh(self.out_ic1(self.up5(u4)))                 # H,    3
+
+        # 掩码解码（从瓶颈 d5 出发，保留全局上下文）
+        mm = self.mask_up_a(torch.cat([self.mask_up_0(d5), con_x4], dim=1))  # H/8,  256
+        mm = self.mask_conv_a(mm)                                     # H/8,  128
+        mm = self.mask_up_b(torch.cat([mm, con_x3], dim=1))         # H/4,  128
+        mm = self.mask_conv_b(mm)                                     # H/4,  64
+        mm = self.mask_up_c(torch.cat([mm, con_x2], dim=1))         # H/2,  64
+        mm = self.mask_conv_c(mm)                                     # H/2,  32
+        mm = self.mask_up_d(torch.cat([mm, con_x1], dim=1))         # H,    32
+        Ms = torch.sigmoid(self.out_ms(mm))
+        Mb = torch.sigmoid(self.out_mb(mm))
 
         return Ms, Mb, Ic4, Ic2, Ic1
 
 
-# Refine网络
 class RefineNet(nn.Module):
-    def __init__(self, in_channels=7):  # 3(Iin)+1(Ms)+3(Ic1) =7
-        super().__init__()
-        # 编码器：下采样
-        self.down1 = DownSample(in_channels, 64)  # 1/2
-        self.down2 = DownSample(64, 128)  # 1/4
-        self.down3 = DownSample(128, 256)  # 1/8
-        self.down4 = DownSample(256, 512)  # 1/16
-        # 解码器：带空洞卷积的上采样
-        self.up1 = DilatedUpSample(512, 256)  # 1/8
-        self.up2 = DilatedUpSample(512, 128)  # 1/4
-        self.up3 = DilatedUpSample(256, 64)  # 1/2
-        self.up4 = DilatedUpSample(128, 64)  # 1/1
-        # 输出层：精细化擦除结果Ire
-        self.out_ire = nn.Conv2d(64, 3, 3, 1, 1)
+    """精细修复网络：EraseNet 风格多尺度空洞卷积，感受野更大。
 
-    def forward(self, x):
-        # 编码器
-        d1 = self.down1(x)
-        d2 = self.down2(d1)
-        d3 = self.down3(d2)
-        d4 = self.down4(d3)
-        # 解码器
-        u1 = self.up1(d4)
-        u1 = torch.cat([u1, d3], dim=1)
-
-        u2 = self.up2(u1)
-        u2 = torch.cat([u2, d2], dim=1)
-
-        u3 = self.up3(u2)
-        u3 = torch.cat([u3, d1], dim=1)
-
-        u4 = self.up4(u3)
-        # 输出
-        Ire = torch.tanh(self.out_ire(u4))
-        return Ire
+    Args:
+        in_channels: 输入通道数 = Iin(3) + Ms(1) + Ic1(3) = 7
+    """
+    def __init__(self, in_channels: int = 7):
+        super().__init__()
+        cnum = 32
+
+        # 编码
+        self.conva = nn.Sequential(
+            nn.Conv2d(in_channels, cnum, 5, padding=2, bias=False),
+            nn.BatchNorm2d(cnum), nn.ReLU(inplace=True))
+        self.down1 = DownSample(cnum,     cnum * 2)                  # H/2, 64
+        self.convc = nn.Sequential(
+            nn.Conv2d(cnum * 2, cnum * 2, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 2), nn.ReLU(inplace=True))
+        self.down2 = DownSample(cnum * 2, cnum * 4)                  # H/4, 128
+        self.conve = nn.Sequential(
+            nn.Conv2d(cnum * 4, cnum * 4, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 4), nn.ReLU(inplace=True))
+        self.convf = nn.Sequential(
+            nn.Conv2d(cnum * 4, cnum * 4, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 4), nn.ReLU(inplace=True))
+
+        # 多尺度空洞卷积（参照 EraseNet astrous_net）
+        self.astrous = nn.Sequential(
+            DilatedConvBlock(cnum * 4, cnum * 4, dilation=2),
+            DilatedConvBlock(cnum * 4, cnum * 4, dilation=4),
+            DilatedConvBlock(cnum * 4, cnum * 4, dilation=8),
+            DilatedConvBlock(cnum * 4, cnum * 4, dilation=16),
+        )
+        self.convk = nn.Sequential(
+            nn.Conv2d(cnum * 4, cnum * 4, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 4), nn.ReLU(inplace=True))
+        self.convl = nn.Sequential(
+            nn.Conv2d(cnum * 4, cnum * 4, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 4), nn.ReLU(inplace=True))
+
+        # 解码（cat 自身跳跃连接）
+        self.up1   = UpSample(cnum * 8, cnum * 2)                    # cat([x,x_c2]) H/4→H/2
+        self.convm = nn.Sequential(
+            nn.Conv2d(cnum * 2, cnum * 2, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum * 2), nn.ReLU(inplace=True))
+        self.up2   = UpSample(cnum * 4, cnum)                        # cat([x,x_c1]) H/2→H
+        self.convn = nn.Sequential(
+            nn.Conv2d(cnum, cnum // 2, 3, padding=1, bias=False),
+            nn.BatchNorm2d(cnum // 2), nn.ReLU(inplace=True),
+            nn.Conv2d(cnum // 2, 3, 3, padding=1))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x    = self.conva(x)
+        x    = self.down1(x)
+        x    = self.convc(x)
+        x_c1 = x                                      # H/2, 64
+        x    = self.down2(x)
+        x    = self.conve(x)
+        x    = self.convf(x)
+        x_c2 = x                                      # H/4, 128
+        x    = self.astrous(x)
+        x    = self.convk(x)
+        x    = self.convl(x)
+        x    = self.up1(torch.cat([x, x_c2], dim=1)) # H/2, 64
+        x    = self.convm(x)
+        x    = self.up2(torch.cat([x, x_c1], dim=1)) # H,   32
+        return torch.tanh(self.convn(x))
 
 
-# 生成器整体
 class Generator(nn.Module):
-    def __init__(self):
+    """完整生成器：CoarseNet → RefineNet → 融合输出。
+
+    Args:
+        cfg: model 子配置字典，含 coarse_in_channels / refine_in_channels / cbam_reduction
+    """
+    def __init__(self, cfg: dict = None):
         super().__init__()
-        self.coarse = CoarseNet()
-        self.refine = RefineNet()
+        if cfg is None:
+            cfg = {'coarse_in_channels': 3, 'refine_in_channels': 7, 'cbam_reduction': 16}
+        self.coarse = CoarseNet(in_channels=cfg['coarse_in_channels'],
+                                cbam_reduction=cfg['cbam_reduction'])
+        self.refine = RefineNet(in_channels=cfg['refine_in_channels'])
 
-    def forward(self, Iin):
-        # Coarse网络输出
+    def forward(self, Iin: torch.Tensor):
         Ms, Mb, Ic4, Ic2, Ic1 = self.coarse(Iin)
-        # 拼接输入：Iin + Ms + Ic1（通道维度）
-        refine_in = torch.cat([Iin, Ms, Ic1], dim=1)
-        # Refine网络输出
-        Ire = self.refine(refine_in)
-        # 融合得到最终结果Icomp
+        Ire   = self.refine(torch.cat([Iin, Ms, Ic1], dim=1))
         Icomp = Ire * Mb + Iin * (1 - Mb)
         return Ms, Mb, Ic4, Ic2, Ic1, Ire, Icomp
-
-
-# 基础判别器块
-class DiscBlock(nn.Module):
-    def __init__(self, in_channels, out_channels, stride=2):
-        super().__init__()
-        self.block = nn.Sequential(
-            nn.Conv2d(in_channels, out_channels, 4, stride, 1, bias=False),
-            nn.BatchNorm2d(out_channels),
-            nn.LeakyReLU(0.2, inplace=True)
-        )
-
-    def forward(self, x):
-        return self.block(x)
-
-
-# Local-Global判别器
-class Discriminator(nn.Module):
-    def __init__(self, in_channels=3):
-        super().__init__()
-        # 全局判别器：输出标量（适配任意尺寸）
-        self.global_disc = nn.Sequential(
-            DiscBlock(in_channels, 64),
-            DiscBlock(64, 128),
-            DiscBlock(128, 256),
-            DiscBlock(256, 512),
-            nn.Conv2d(512, 1, 4, 1, 0, bias=False)  # [B,1,1,1]
-        )
-        # 局部判别器：输出特征图（用于掩码加权）
-        self.local_disc = nn.Sequential(
-            DiscBlock(in_channels, 64),
-            DiscBlock(64, 128),
-            DiscBlock(128, 256),
-            DiscBlock(256, 512),
-            nn.Conv2d(512, 1, 1, 1, 0, bias=False)  # [B,1,H',W']
-        )
-
-    def forward(self, x, local_mask=None):
-        """
-        :param x: [B,3,H,W] 归一化到[-1,1]的图像
-        :param local_mask: [B,1,H,W] 真实文本掩码（训练时必需）
-        :return:
-            global_score: [B,1,1,1] 全局判别logits
-            local_score: [B,1,H',W'] 掩码加权后的局部判别logits
-        """
-        global_score = self.global_disc(x)
-        local_score = 0
-
-        if local_mask is not None:
-            # 1. 计算局部判别特征图
-            local_feat = self.local_disc(x)  # [B,1,H',W']
-            # 2. 将local_mask下采样到local_feat的尺寸
-            _, _, h_feat, w_feat = local_feat.shape
-            local_mask_scaled = F.interpolate(local_mask, size=(h_feat, w_feat), mode='nearest')
-            # 3. 关键修复：用掩码加权局部特征（仅关注文本区域）
-            local_score = local_feat * local_mask_scaled
-
-        return global_score, local_score
-
-
-# 加载预训练VGG16（用于感知/风格损失）
-class VGG16Feature(nn.Module):
-    def __init__(self):
-        super().__init__()
-        vgg = vgg16(pretrained=True).features
-        self.feat1 = nn.Sequential(*vgg[:5])  # 第1个池化层前
-        self.feat2 = nn.Sequential(*vgg[5:10])  # 第2个池化层前
-        self.feat3 = nn.Sequential(*vgg[10:17])  # 第3个池化层前
-        # 冻结参数
-        for param in self.parameters():
-            param.requires_grad = False
-
-            # 【关键修正】ImageNet归一化参数
-        # self.mean = torch.tensor([0.485, 0.456, 0.406]).view(1, 3, 1, 1).cuda()
-        # self.std = torch.tensor([0.229, 0.224, 0.225]).view(1, 3, 1, 1).cuda()
-        self.register_buffer('mean', torch.tensor([0.485, 0.456, 0.406]).view(1, 3, 1, 1))
-        self.register_buffer('std', torch.tensor([0.229, 0.224, 0.225]).view(1, 3, 1, 1))
-
-    def forward(self, x):
-        # x: [B,3,H,W] 归一化到ImageNet标准
-        # x: [-1,1] → [0,1] → ImageNet归一化
-        x = (x + 1) / 2.0  # [-1,1] → [0,1]
-        x = F.interpolate(x, size=(224, 224), mode='bilinear', align_corners=False)
-        x = (x - self.mean) / self.std
-        f1 = self.feat1(x)
-        f2 = self.feat2(f1)
-        f3 = self.feat3(f2)
-        return [f1, f2, f3]
-
-
-# Gram矩阵计算
-def gram_matrix(x):
-    b, c, h, w = x.shape
-    x = x.view(b, c, h * w)
-    return torch.bmm(x, x.transpose(1, 2)) / (c * h * w)
-
-
-# 全量Loss计算类
-class EnsExamLoss(nn.Module):
-    def __init__(self, loss_config=None):
-        super().__init__()
-        if loss_config is None:
-            loss_config = {}
-        self.vgg_feat = VGG16Feature()
-        self.lambda_n = loss_config.get('lambda_n', [5, 6, 8, 10])
-        self.beta_n = loss_config.get('beta_n', [0.8, 0.8, 0.8, 2])
-        self.lambda_lr = loss_config.get('lambda_lr', 1.0)
-        self.lambda_p = loss_config.get('lambda_p', 0.05)
-        self.lambda_style = loss_config.get('lambda_style', 120)
-        self.lambda_sn = loss_config.get('lambda_sn', 1)
-        self.lambda_b = loss_config.get('lambda_b', 0.4)
-
-    def sn_loss(self, Ms, Ms_gt):
-        """SN损失"""
-        Ms = Ms.float()
-        Ms_gt = Ms_gt.float()
-
-        # 1. 计算每个样本的 GT 字迹总面积 (Batch 维度)
-        # shape: [B]
-        sum_gt = torch.sum(Ms_gt, dim=tuple(range(1, Ms.dim())))
-
-        # 2. 创建掩码：只选择 GT 中有字迹的样本 (阈值设为 1，避免浮点误差)
-        # 如果 sum_gt > 0，则 valid_mask 为 1，否则为 0
-        valid_mask = (sum_gt > 1.0).float()
-
-        # 如果整个 Batch 都没有字迹，直接返回 0，避免除零或无效计算
-        if valid_mask.sum() == 0:
-            # 返回与输入无关的零梯度，但保持可训练
-            return Ms.sum() * 0  # 利用输入创建，保持计算图连接
-
-        # 3. 计算 L1 误差总和 [B]
-        l1_sum = torch.sum(torch.abs(Ms - Ms_gt), dim=tuple(range(1, Ms.dim())))
-
-        # 4. 计算归一化分母 [B]
-        sum_pred = torch.sum(Ms, dim=tuple(range(1, Ms.dim())))
-        normalization = torch.min(sum_pred, sum_gt)
-
-        # 5. 计算单样本 Loss
-        # 注意：这里 epsilon 可以改回 1e-6，因为 valid_mask 已经保证了 sum_gt 不为 0
-        loss_batch = l1_sum / (normalization + 1e-6)
-
-        # 6. 只对有字迹的样本求平均
-        # 乘以 valid_mask 将无字样本 loss 置零，除以 valid_mask.sum() 保证梯度尺度稳定
-        L_sn = (loss_batch * valid_mask).sum() / (valid_mask.sum() + 1e-6)
-
-        return L_sn
-
-    def block_loss(self, Mb, Mb_gt):
-        # 逐样本计算
-        intersection = (Mb * Mb_gt).sum(dim=[1, 2, 3])  # [B]
-        mb_sq = (Mb ** 2).sum(dim=[1, 2, 3])  # [B]
-        mbgt_sq = (Mb_gt ** 2).sum(dim=[1, 2, 3])  # [B]
-
-        dice = (2 * intersection) / (mb_sq + mbgt_sq + 1e-8)  # [B]
-        return (1 - dice).mean()  # 平均所有样本
-
-    def lr_loss(self, Iouts, Igt_list, Mb_gt):
-        """LR损失（修复多尺度掩码维度匹配问题）"""
-        lr_loss = 0.0
-        for i, (Iout, Igt) in enumerate(zip(Iouts, Igt_list)):
-            # 关键修复：将Mb_gt下采样到当前Iout/Igt的尺度
-            # 获取当前Iout的高/宽
-            _, _, h, w = Iout.shape
-            # 下采样Mb_gt到(h, w)，保持通道数不变
-            Mb_gt_scaled = F.interpolate(Mb_gt, size=(h, w), mode='nearest')
-
-            # 文本区域损失（使用下采样后的掩码）
-            loss_text = F.l1_loss(Iout * Mb_gt_scaled, Igt * Mb_gt_scaled, reduction='mean')
-            # 非文本区域损失（使用下采样后的掩码）
-            loss_nontext = F.l1_loss(Iout * (1 - Mb_gt_scaled), Igt * (1 - Mb_gt_scaled), reduction='mean')
-            # 加权求和
-            lr_loss += self.lambda_n[i] * loss_text + self.beta_n[i] * loss_nontext
-        return lr_loss
-
-    def perceptual_loss(self, I_list, Igt):
-        """感知损失"""
-        per_loss = 0.0
-        gt_feats = self.vgg_feat(Igt)
-        for I in I_list:
-            I_feats = self.vgg_feat(I)
-            for f1, f2 in zip(I_feats, gt_feats):
-                per_loss += F.l1_loss(f1, f2, reduction='mean')
-        return per_loss
-
-    def style_loss(self, I_list, Igt):
-        style_loss = 0.0
-        gt_feats = self.vgg_feat(Igt)
-        for I in I_list:
-            I_feats = self.vgg_feat(I)
-            for f1, f2 in zip(I_feats, gt_feats):
-                b, c, h, w = f1.shape
-                gram1 = gram_matrix(f1)  # [B, C, C]
-                gram2 = gram_matrix(f2)  # [B, C, C]
-                # 关键：除以 H*W*C 归一化
-                norm = h * w * c
-                style_loss += F.l1_loss(gram1, gram2, reduction='mean') / norm
-        return style_loss
-
-    # 判别器损失（训练D时用）
-    @staticmethod
-    def hinge_loss_D(real_score, fake_score):
-        loss_real = torch.mean(F.relu(1.0 - real_score))
-        loss_fake = torch.mean(F.relu(1.0 + fake_score))
-        return loss_real + loss_fake
-
-    # 生成器损失（训练G时用）
-    @staticmethod
-    def hinge_loss_G(fake_score):
-        return -torch.mean(F.relu(1.0 + fake_score))  # 注意：ReLU(1+score)
-
-    def forward(self, gen_out, gt, disc_score):
-        """
-        总损失计算
-        :param gen_out: 生成器输出 (Ms, Mb, Ic4, Ic2, Ic1, Ire, Icomp)
-        :param gt: 标签 (Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt)
-        :param disc_score: 判别器分数 (global_score, local_score)
-        :return: 总损失，各分项损失
-        """
-        Ms, Mb, Ic4, Ic2, Ic1, Ire, Icomp = gen_out
-        Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt = gt
-        global_score, local_score = disc_score
-
-        # 各分项损失
-        L_sn = self.sn_loss(Ms, Ms_gt) * self.lambda_sn
-        L_block = self.block_loss(Mb, Mb_gt) * self.lambda_b
-        L_lr = self.lr_loss([Ic4, Ic2, Ic1, Ire], [Igt4, Igt2, Igt1, Igt], Mb_gt) * self.lambda_lr
-        L_per = self.perceptual_loss([Ire, Icomp], Igt) * self.lambda_p
-        L_style = self.style_loss([Ire, Icomp], Igt) * self.lambda_style
-
-        L_adv_global = self.hinge_loss_G(global_score)
-        L_adv_local = self.hinge_loss_G(local_score)
-        L_adv = (L_adv_global + L_adv_local) / 2
-
-        # 总损失
-        L_total = L_adv + L_lr + L_per + L_style + L_sn + L_block
-        return L_total, [L_adv, L_lr, L_per, L_style, L_sn, L_block]
\ No newline at end of file
diff --git a/backend/models/test_erase.py b/backend/models/test_erase.py
index 59c13aa4..e0eee131 100644
--- a/backend/models/test_erase.py
+++ b/backend/models/test_erase.py
@@ -9,8 +9,7 @@
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
 
 import torch
-torch.backends.cudnn.deterministic = True
-torch.backends.cudnn.benchmark = False
+torch.backends.cudnn.enabled = False   # cuDNN v8 frontend 与 A30+PyTorch2.1 不兼容
 from pathlib import Path
 from PIL import Image
 import numpy as np
@@ -27,7 +26,7 @@
     sys.exit(1)
 
 # ── 路径配置 ──────────────────────────────────────────
-MODEL_PATH = Path(__file__).parent.parent / "runtime_data" / "models" / "latest.pth"
+MODEL_PATH = Path(__file__).parent / "weight" / "best.pth"
 OUTPUT_DIR = Path(__file__).parent.parent / "runtime_data" / "erased"
 OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
diff --git a/backend/models/train.py b/backend/models/train.py
deleted file mode 100644
index 9baa621a..00000000
--- a/backend/models/train.py
+++ /dev/null
@@ -1,556 +0,0 @@
-# 设备
-import math
-import os
-import argparse
-
-import numpy as np
-import torch
-from torch import optim
-from torch.utils.data import Dataset, DataLoader
-from torchvision.transforms import transforms
-import torch.nn.functional as F
-from tqdm import tqdm
-
-from src.adcj.EnsExam.model import Generator, Discriminator, EnsExamLoss
-from src.adcj.EnsExam.config import config
-
-import os
-import torch
-import numpy as np
-import cv2
-from torch.utils.data import Dataset
-from torchvision import transforms
-import torch.nn.functional as F
-
-
-def generate_mask_from_pair(Iin, Igt, threshold=20, debug=False):
-    """
-    单块（512x512）软笔画掩码生成 - 修正版本
-    :param Iin: 单块图像 (H,W,3) RGB 0-255
-    :param Igt: 单块GT图 (H,W,3) RGB 0-255
-    :param threshold: 差异阈值，默认20
-    :param debug: 是否返回中间结果用于调试
-    :return: Ms_gt(软掩码), Mb_gt(基础掩码), [debug_dict]
-    """
-    H, W = Iin.shape[:2]
-
-    # ========== 1. 生成粗笔画掩码 ==========
-    # 使用int16避免溢出，计算RGB三通道的平均差异
-    diff = np.abs(Iin.astype(np.int16) - Igt.astype(np.int16)).mean(axis=-1)
-    # print(f"Diff统计: min={diff.min()}, max={diff.max()}, 90%={np.percentile(diff, 90)}")
-    coarse_mask = (diff > threshold).astype(np.uint8)
-
-    # ========== 2. 形态学去噪 ==========
-    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
-    # 开运算去除散点噪声 + 闭运算连接断裂笔画
-    denoised_mask = cv2.morphologyEx(coarse_mask, cv2.MORPH_OPEN, kernel)
-    denoised_mask = cv2.morphologyEx(denoised_mask, cv2.MORPH_CLOSE, kernel)
-
-    # ========== 3. 生成 Mb_gt（基础文本块掩码） ==========
-    # 对去噪后的掩码适度膨胀，作为文本区域指导
-    Mb_gt = cv2.dilate(denoised_mask, kernel, iterations=2).astype(np.float32)
-
-    # ========== 4. 生成软笔画掩码 Ms_gt ==========
-    # 4.1 骨架：笔画收缩1像素（论文明确要求）
-    skeleton = cv2.erode(denoised_mask, kernel, iterations=1)
-
-    # 4.2 外边界区域：原始笔画扩张5像素
-    kernel_large = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
-    outer_region = cv2.dilate(denoised_mask, kernel_large, iterations=5)
-
-    # 4.3 距离变换：计算outer_region内每个像素到边缘的最短距离
-    # 输入要求：uint8, 0=背景, >0=前景；输出：float32，单位像素
-    dist_map = cv2.distanceTransform(outer_region, cv2.DIST_L2, cv2.DIST_MASK_PRECISE)
-
-    # 4.4 SAF 参数 (论文明确值)
-    alpha = 3.0
-    L = 5.0
-    exp_neg_alpha = np.exp(-alpha)
-    C = (1 + exp_neg_alpha) / (1 - exp_neg_alpha + 1e-8)  # +eps防除零
-
-    # 4.5 向量化计算软掩码（避免循环，提速100x）
-    Ms_gt = np.zeros((H, W), dtype=np.float32)
-
-    # 区域划分mask
-    mask_skeleton = skeleton > 0  # 骨架区域 → 值=1
-    mask_outer = outer_region > 0  # 扩张区域（含骨架）
-    mask_middle = mask_outer & (~mask_skeleton)  # 中间环形区域 → 用SAF
-
-    # (a) 骨架区域强制=1
-    Ms_gt[mask_skeleton] = 1.0
-
-    # (b) 中间区域应用SAF衰减
-    if np.any(mask_middle):
-        D = np.clip(dist_map[mask_middle], 0, L)  # 距离截断到[0, L]
-        # SAF公式: C * (2/(1+exp(-α*D/L)) - 1)
-        exp_term = np.exp(-alpha * D / L)
-        saf_vals = C * (2.0 / (1.0 + exp_term + 1e-8) - 1.0)
-        Ms_gt[mask_middle] = np.clip(saf_vals, 0.0, 1.0)
-
-    # (c) 外边界及以外区域保持=0（已初始化为0）
-
-    # ========== 5. 调试信息（可选） ==========
-    if debug:
-        debug_info = {
-            'coarse_mask': coarse_mask,
-            'denoised_mask': denoised_mask,
-            'skeleton': skeleton,
-            'outer_region': outer_region,
-            'dist_map': dist_map,
-            'mask_middle': mask_middle.astype(np.uint8) * 255
-        }
-        return Ms_gt, Mb_gt, debug_info
-
-    return Ms_gt, Mb_gt
-
-
-# -------------------------- 真实数据集加载类（核心修复：统一图像尺寸） --------------------------
-# class EnsExamRealDataset(Dataset):
-#     """
-#     适配“仅含原始图+擦除GT图”的数据集加载类
-#     ✅ 修复：所有图像读取后先resize到img_size，保证批量尺寸一致
-#     """
-#
-#     def __init__(self, data_root, img_size=512, is_train=True):
-#         self.data_root = data_root
-#         self.img_size = img_size  # 统一目标尺寸（512×512）
-#         self.is_train = is_train
-#
-#         # 1. 定义图像预处理：归一化到[-1,1]（匹配GAN训练）
-#         self.img_transform = transforms.Compose([
-#             transforms.ToTensor(),  # HWC→CHW，0-255→0-1
-#             transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])  # 0-1→[-1,1]
-#         ])
-#
-#         # 2. 加载文件列表（保证images和gt目录下文件同名）
-#         split = "train" if is_train else "test"
-#         self.img_dir = os.path.join(data_root, split, "all_images")
-#         self.gt_dir = os.path.join(data_root, split, "all_labels")
-#
-#         # 过滤有效文件（仅保留png/jpg/jpeg）
-#         self.file_names = [
-#             f for f in os.listdir(self.img_dir)
-#             if f.endswith((".png", ".jpg", ".jpeg")) and os.path.exists(os.path.join(self.gt_dir, f))
-#         ]
-#         assert len(self.file_names) > 0, f"未找到匹配的图像文件！检查{self.img_dir}和{self.gt_dir}目录"
-#
-#     def __len__(self):
-#         return len(self.file_names)
-#
-#     def __getitem__(self, idx):
-#         # 1. 加载原始图和擦除GT图
-#         file_name = self.file_names[idx]
-#         img_path = os.path.join(self.img_dir, file_name)
-#         gt_path = os.path.join(self.gt_dir, file_name)
-#
-#         # 读取图像（BGR→RGB，避免OpenCV默认BGR格式问题）
-#         Iin = cv2.imread(img_path)[:, :, ::-1]  # (H,W,3)，0-255，RGB
-#         Igt = cv2.imread(gt_path)[:, :, ::-1]  # (H,W,3)，0-255，RGB
-#
-#         # ✅ 核心修复1：先统一resize到img_size×img_size（所有样本尺寸一致）
-#         Iin = cv2.resize(Iin, (self.img_size, self.img_size), interpolation=cv2.INTER_LINEAR)
-#         Igt = cv2.resize(Igt, (self.img_size, self.img_size), interpolation=cv2.INTER_LINEAR)
-#
-#         # 2. 自动生成Ms_gt（软笔画掩码）、Mb_gt（文本块掩码）
-#         # ✅ 核心修复2：传入已resize的图像，函数内不再重复resize
-#         Ms_gt_np, Mb_gt_np = generate_mask_from_pair(Iin, Igt, threshold=20)
-#
-#         # 3. 图像预处理（归一化到[-1,1]）
-#         Iin = self.img_transform(Iin)  # (3,512,512)，[-1,1]
-#         Igt = self.img_transform(Igt)  # (3,512,512)，[-1,1]
-#
-#         # 4. 掩码预处理（归一化到[0,1]，扩展通道维度）
-#         Ms_gt = torch.from_numpy(Ms_gt_np).unsqueeze(0).float()  # (1,512,512)，[0,1]
-#         Mb_gt = torch.from_numpy(Mb_gt_np).unsqueeze(0).float()  # (1,512,512)，[0,1]
-#
-#         # 5. 生成多尺度擦除GT（1/4, 1/2, 1/1）
-#         # Igt4：512→128，Igt2：512→256，Igt1：原尺寸
-#         Igt4 = F.interpolate(Igt.unsqueeze(0), size=(self.img_size // 4, self.img_size // 4), mode='bilinear',
-#                              align_corners=False).squeeze(0)
-#         Igt2 = F.interpolate(Igt.unsqueeze(0), size=(self.img_size // 2, self.img_size // 2), mode='bilinear',
-#                              align_corners=False).squeeze(0)
-#         Igt1 = Igt  # 1:1尺度
-#
-#         # 返回顺序：Iin, Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt
-#         return Iin, Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt
-
-class EnsExamRealDataset(Dataset):
-    """
-    适配“仅含原始图 + 擦除 GT 图”的数据集加载类
-    ✅ 修改：不再 resize 原图，而是将大图裁剪为多个 512x512 的样本块
-    """
-
-    def __init__(self, data_root, img_size=512, is_train=True, overlap=0):
-        """
-        :param data_root: 数据集根目录
-        :param img_size: 裁剪块尺寸 (默认 512)
-        :param is_train: 是否训练模式
-        :param overlap: 裁剪重叠像素 (默认 0，即不重叠；训练时可设为 128 增加数据量)
-        """
-        self.data_root = data_root
-        self.img_size = img_size
-        self.is_train = is_train
-        self.overlap = overlap
-
-        # 1. 定义图像预处理：归一化到 [-1,1]
-        self.img_transform = transforms.Compose([
-            transforms.ToTensor(),  # HWC→CHW, 0-255→0-1
-            transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
-        ])
-
-        # 2. 获取文件路径
-        split = "train" if is_train else "test"
-        self.img_dir = os.path.join(data_root, split, "all_images")
-        self.gt_dir = os.path.join(data_root, split, "all_labels")
-
-        # 3. ✅ 核心修改：构建“样本索引列表”
-        # 不再是一对一 (1 图=1 样本)，而是一对多 (1 图=N 个样本块)
-        self.patch_index_map = []
-
-        valid_extensions = (".png", ".jpg", ".jpeg")
-        all_files = [f for f in os.listdir(self.img_dir) if f.endswith(valid_extensions)]
-
-        print(f"🔍 正在扫描数据集并构建裁剪索引 (Overlap={overlap})...")
-        for fname in all_files:
-            if not os.path.exists(os.path.join(self.gt_dir, fname)):
-                continue
-
-            img_path = os.path.join(self.img_dir, fname)
-            gt_path = os.path.join(self.gt_dir, fname)
-
-            # 读取尺寸 (仅读取头信息，不加载像素，速度快)
-            # 如果 cv2.imread 太慢，可用 imageio 或 PIL 获取 size，这里为了兼容直接用 cv2
-            # 注意：这里必须加载一次获取 H,W，或者用 cv2.imread + IMREAD_UNCHANGED
-            img_temp = cv2.imread(img_path)
-            if img_temp is None: continue
-            H, W = img_temp.shape[:2]
-            del img_temp  # 释放内存
-
-            # 计算步长
-            step = self.img_size - self.overlap
-            if step <= 0: step = 1  # 防止 overlap 过大导致步长为 0
-
-            # 计算该图能切多少块 (覆盖全图，边缘不足补黑)
-            # 使用 math.ceil 确保覆盖整张图
-            num_h = math.ceil((H - self.overlap) / step) if H > self.img_size else 1
-            num_w = math.ceil((W - self.overlap) / step) if W > self.img_size else 1
-
-            # 如果原图小于 512，也作为一个样本 (后续会 padding)
-            if H <= self.img_size and W <= self.img_size:
-                num_h, num_w = 1, 1
-
-            for i in range(num_h):
-                for j in range(num_w):
-                    # 计算裁剪坐标
-                    y1 = i * step
-                    x1 = j * step
-
-                    # 如果是最后一块，确保覆盖到边缘
-                    y2 = min(y1 + self.img_size, H)
-                    x2 = min(x1 + self.img_size, W)
-
-                    # 如果是第一块且图很小，直接取全图
-                    if H <= self.img_size: y1, y2 = 0, H
-                    if W <= self.img_size: x1, x2 = 0, W
-
-                    # 记录样本信息：(图路径，GT 路径，裁剪坐标，是否需要 padding)
-                    need_pad_h = (y2 - y1) < self.img_size
-                    need_pad_w = (x2 - x1) < self.img_size
-
-                    self.patch_index_map.append({
-                        'img_path': img_path,
-                        'gt_path': gt_path,
-                        'y1': y1, 'y2': y2,
-                        'x1': x1, 'x2': x2,
-                        'pad_h': need_pad_h,
-                        'pad_w': need_pad_w
-                    })
-
-        print(f"✅ 索引构建完成：共 {len(self.patch_index_map)} 个样本块 (来自 {len(all_files)} 张图)")
-        assert len(self.patch_index_map) > 0, "未找到有效样本！"
-
-    def __len__(self):
-        return len(self.patch_index_map)
-
-    def __getitem__(self, idx):
-        # 1. 获取当前样本的裁剪信息
-        info = self.patch_index_map[idx]
-
-        # print(f"🔎 Loading: {info['img_path']}")
-        # print(f"   Exists: {os.path.exists(info['img_path'])}")
-
-        # 2. 加载完整原图 (RGB)
-        Iin_full = cv2.imread(info['img_path'])[:, :, ::-1]
-        Igt_full = cv2.imread(info['gt_path'])[:, :, ::-1]
-
-        # 3. ✅ 核心修改：根据坐标裁剪
-        Iin = Iin_full[info['y1']:info['y2'], info['x1']:info['x2']]
-        Igt = Igt_full[info['y1']:info['y2'], info['x1']:info['x2']]
-
-        Iin = np.ascontiguousarray(Iin)
-        Igt = np.ascontiguousarray(Igt)
-
-        # 4. ✅ 边缘填充 (确保输入模型的都是 512x512)
-        if info['pad_h'] or info['pad_w']:
-            pad_h = self.img_size - Iin.shape[0]
-            pad_w = self.img_size - Iin.shape[1]
-            # 使用 REPLICATE 填充比 CONSTANT(黑边) 更好，减少边界伪影
-            Iin = cv2.copyMakeBorder(Iin, 0, pad_h, 0, pad_w, cv2.BORDER_REPLICATE)
-            Igt = cv2.copyMakeBorder(Igt, 0, pad_h, 0, pad_w, cv2.BORDER_REPLICATE)
-
-        # 5. 生成掩码 (输入已经是 512x512)
-        # 注意：这里调用的是单块生成函数，不是滑动拼接函数
-        Ms_gt_np, Mb_gt_np = generate_mask_from_pair(Iin, Igt, threshold=20)
-
-        # 6. 图像预处理 (归一化)
-        Iin = self.img_transform(Iin)  # (3, 512, 512)
-        Igt = self.img_transform(Igt)  # (3, 512, 512)
-
-        # 7. 掩码转 Tensor
-        Ms_gt = torch.from_numpy(Ms_gt_np).unsqueeze(0).float()
-        Mb_gt = torch.from_numpy(Mb_gt_np).unsqueeze(0).float()
-
-        # 8. 生成多尺度 GT (1/4, 1/2, 1/1)
-        Igt_unsqueeze = Igt.unsqueeze(0)
-        Igt4 = F.interpolate(Igt_unsqueeze, size=(self.img_size // 4, self.img_size // 4),
-                             mode='bilinear', align_corners=False).squeeze(0)
-        Igt2 = F.interpolate(Igt_unsqueeze, size=(self.img_size // 2, self.img_size // 2),
-                             mode='bilinear', align_corners=False).squeeze(0)
-        Igt1 = Igt
-
-        # 返回：Iin, Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt
-        return Iin, Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt
-
-
-# ====================== 3. 核心训练函数 ======================
-def train_ensexam(
-        epochs=None,
-        batch_size=None,
-        lr=None,
-        data_root=None,
-        img_size=512,
-        save_dir=None,
-        resume_path=None,
-        resume=False,
-        num_workers=None,
-        device=None,
-        use_wandb=None):
-    train_cfg = config.train
-    loss_cfg = config.loss
-    wandb_cfg = config.wandb
-
-    epochs = epochs if epochs is not None else train_cfg.get('epochs', 100)
-    batch_size = batch_size if batch_size is not None else train_cfg.get('batch_size', 4)
-    lr = lr if lr is not None else train_cfg.get('lr', 0.0001)
-    data_root = data_root if data_root is not None else train_cfg.get('data_root', './data')
-    save_dir = save_dir if save_dir is not None else train_cfg.get('save_dir', './ensexam_checkpoints')
-    resume_path = resume_path if resume_path is not None else train_cfg.get('resume_path',
-                                                                            './ensexam_checkpoints/latest.pth')
-    num_workers = num_workers if num_workers is not None else train_cfg.get('num_workers', 4)
-    device = device if device is not None else train_cfg.get('device', 'cuda' if torch.cuda.is_available() else 'cpu')
-    use_wandb = use_wandb if use_wandb is not None else wandb_cfg.get('enabled', True)
-    # 0. 初始化目录
-    os.makedirs(save_dir, exist_ok=True)
-
-    wandb_run = None
-    if use_wandb:
-        try:
-            import wandb
-            wandb_cfg = config.wandb
-            if wandb_cfg.get('offline', True):
-                os.environ['WANDB_MODE'] = 'offline'
-                wandb_dir = wandb_cfg.get('dir', './wandb')
-                os.makedirs(wandb_dir, exist_ok=True)
-
-            wandb_run = wandb.init(
-                project=wandb_cfg.get('project', 'ensexam'),
-                name=wandb_cfg.get('name'),
-                entity=wandb_cfg.get('entity'),
-                dir=wandb_cfg.get('dir', './wandb'),
-                config={
-                    'train': train_cfg,
-                    'loss': loss_cfg
-                }
-            )
-            print(f"WandB 初始化成功 (离线模式: {wandb_cfg.get('offline', True)})")
-        except ImportError:
-            print("Warning: wandb 未安装，跳过 wandb 记录")
-        except Exception as e:
-            print(f"Warning: wandb 初始化失败: {e}")
-
-    # 1. 数据加载
-    train_dataset = EnsExamRealDataset(data_root=data_root)
-    train_loader = DataLoader(
-        train_dataset,
-        batch_size=batch_size,
-        shuffle=True,
-        num_workers=0,  # Windows下建议设为0，避免多进程问题
-        drop_last=True
-    )
-
-    # 2. 模型初始化
-    G = Generator().to(device)
-    D = Discriminator().to(device)
-    criterion = EnsExamLoss(loss_cfg).to(device)
-
-    # 3. 优化器
-    optimizer_G = optim.Adam(G.parameters(), lr=lr, betas=(0.5, 0.9))
-    optimizer_D = optim.Adam(D.parameters(), lr=lr, betas=(0.5, 0.9))
-
-    # 4. 断点续训
-    start_epoch = 0
-    if resume and os.path.exists(resume_path):
-        checkpoint = torch.load(resume_path, map_location=device)
-        G.load_state_dict(checkpoint['G_state_dict'])
-        D.load_state_dict(checkpoint['D_state_dict'])
-        optimizer_G.load_state_dict(checkpoint['optimizer_G'])
-        optimizer_D.load_state_dict(checkpoint['optimizer_D'])
-        start_epoch = checkpoint['epoch']
-        print(f"断点续训：从第{start_epoch}Epoch恢复")
-
-    # 5. 训练循环
-    G.train()
-    D.train()
-    for epoch in range(start_epoch, epochs):
-        epoch_loss_G = 0.0  # 生成器总损失
-        epoch_loss_D = 0.0  # 判别器总损失
-        epoch_loss_parts = {
-            'adv': 0.0,
-            'lr': 0.0,
-            'per': 0.0,
-            'style': 0.0,
-            'sn': 0.0,
-            'block': 0.0
-        }
-        pbar = tqdm(train_loader, desc=f"Epoch {epoch + 1}/{epochs}")
-
-        for batch_idx, (Iin, Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt) in enumerate(pbar):
-            # 数据移到设备
-            Iin = Iin.to(device)
-            Ms_gt = Ms_gt.to(device)
-            Mb_gt = Mb_gt.to(device)
-            Igt4 = Igt4.to(device)
-            Igt2 = Igt2.to(device)
-            Igt1 = Igt1.to(device)
-            Igt = Igt.to(device)
-            gt = (Ms_gt, Mb_gt, Igt4, Igt2, Igt1, Igt)
-
-            # ---------------------- 步骤1：训练判别器D ----------------------
-            optimizer_D.zero_grad()
-
-            # 1.1 真实图像判别
-            real_global, real_local = D(Igt, Mb_gt)
-            # 1.2 生成图像判别
-            gen_out = G(Iin)
-            Icomp = gen_out[-1]  # 取最后一个输出Icomp
-            fake_global, fake_local = D(Icomp.detach(), Mb_gt)
-            # 1.3 计算D损失（使用Model中的hinge_loss_D）
-            loss_D_global = EnsExamLoss.hinge_loss_D(real_global, fake_global)
-            loss_D_local = EnsExamLoss.hinge_loss_D(real_local, fake_local)
-            loss_D = (loss_D_global + loss_D_local) / 2
-
-            loss_D.backward()
-            optimizer_D.step()
-
-            # ---------------------- 步骤2：训练生成器G ----------------------
-            optimizer_G.zero_grad()
-
-            # 2.1 重新计算生成图像的判别分数
-            fake_global, fake_local = D(Icomp, Mb_gt)
-            disc_score = (fake_global, fake_local)
-            # 2.2 计算G的全量损失（Model中的EnsExamLoss.forward）
-            loss_G, loss_parts = criterion(gen_out, gt, disc_score)
-
-            loss_G.backward()
-            optimizer_G.step()
-
-            # ---------------------- 步骤3：日志记录 ----------------------
-            epoch_loss_G += loss_G.item()
-            epoch_loss_D += loss_D.item()
-            epoch_loss_parts['adv'] += loss_parts[0].item()
-            epoch_loss_parts['lr'] += loss_parts[1].item()
-            epoch_loss_parts['per'] += loss_parts[2].item()
-            epoch_loss_parts['style'] += loss_parts[3].item()
-            epoch_loss_parts['sn'] += loss_parts[4].item()
-            epoch_loss_parts['block'] += loss_parts[5].item()
-
-            # 进度条显示
-            pbar.set_postfix({
-                'Loss_D': f"{loss_D.item():.4f}",
-                'Loss_G': f"{loss_G.item():.4f}",
-                'Loss_G_adv': f"{loss_parts[0].item():.4f}",
-                'Loss_G_lr': f"{loss_parts[1].item():.4f}"
-            })
-            if wandb_run is not None:
-                wandb.log({
-                    'batch_loss_G': loss_G.item(),
-                    'batch_loss_D': loss_D.item(),
-                    'batch_loss_G_adv': loss_parts[0].item(),
-                    'batch_loss_G_lr': loss_parts[1].item(),
-                    'batch_loss_G_per': loss_parts[2].item(),
-                    'batch_loss_G_style': loss_parts[3].item(),
-                    'batch_loss_G_sn': loss_parts[4].item(),
-                    'batch_loss_G_block': loss_parts[5].item()
-                })
-
-        # ---------------------- 步骤4：Epoch结束处理 ----------------------
-        # 计算Epoch平均损失
-        avg_loss_G = epoch_loss_G / len(train_loader)
-        avg_loss_D = epoch_loss_D / len(train_loader)
-        avg_loss_parts = {k: v / len(train_loader) for k, v in epoch_loss_parts.items()}
-        print(f"Epoch {epoch + 1} 平均损失：Loss_G={avg_loss_G:.4f}, Loss_D={avg_loss_D:.4f}")
-        if wandb_run is not None:
-            wandb.log({
-                'epoch': epoch + 1,
-                'avg_loss_G': avg_loss_G,
-                'avg_loss_D': avg_loss_D
-            })
-
-        # 保存模型（每5个Epoch保存一次，同时保存最新模型）
-        if (epoch + 1) % 5 == 0 or epoch == epochs - 1:
-            checkpoint = {
-                'epoch': epoch + 1,
-                'G_state_dict': G.state_dict(),
-                'D_state_dict': D.state_dict(),
-                'optimizer_G': optimizer_G.state_dict(),
-                'optimizer_D': optimizer_D.state_dict(),
-                'avg_loss_G': avg_loss_G,
-                'avg_loss_D': avg_loss_D
-            }
-            # 保存最新模型（用于断点续训）
-            torch.save(checkpoint, os.path.join(save_dir, 'latest.pth'))
-            # 保存Epoch模型
-            torch.save(checkpoint, os.path.join(save_dir, f'ensexam_epoch_{epoch + 1}.pth'))
-            print(f"模型已保存：{os.path.join(save_dir, f'ensexam_epoch_{epoch + 1}.pth')}")
-
-    if wandb_run is not None:
-        wandb.finish()
-    print("训练完成！")
-
-
-# ====================== 4. 启动训练 ======================
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description='EnsExam 训练')
-    parser.add_argument('--epochs', type=int, default=None, help='训练轮数')
-    parser.add_argument('--batch_size', type=int, default=None, help='批次大小')
-    parser.add_argument('--lr', type=float, default=None, help='学习率')
-    parser.add_argument('--device', type=str, default=None, help='设备 (cuda/cpu)')
-    parser.add_argument('--data_root', type=str, default=None, help='数据集路径')
-    parser.add_argument('--save_dir', type=str, default=None, help='模型保存目录')
-    parser.add_argument('--resume', action='store_true', help='断点续训')
-    parser.add_argument('--no_wandb', action='store_true', help='禁用 wandb')
-    args = parser.parse_args()
-
-    use_wandb = not args.no_wandb
-    device = args.device
-
-    train_ensexam(
-        epochs=args.epochs,
-        batch_size=args.batch_size,
-        lr=args.lr,
-        device=device,
-        data_root=args.data_root,
-        save_dir=args.save_dir,
-        resume=args.resume,
-        use_wandb=use_wandb
-    )
\ No newline at end of file
diff --git a/backend/models/weight/best.pth b/backend/models/weight/best.pth
new file mode 100644
index 00000000..61723552
--- /dev/null
+++ b/backend/models/weight/best.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ce399f0447fdc016bbe3957559264735a5b9e5de30725ca2ed4d66699fd3f47
+size 362841171
diff --git a/backend/package-lock.json b/backend/package-lock.json
new file mode 100644
index 00000000..dfb18f11
--- /dev/null
+++ b/backend/package-lock.json
@@ -0,0 +1,6 @@
+{
+  "name": "backend",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {}
+}
diff --git a/backend/routes/__init__.py b/backend/routes/__init__.py
new file mode 100644
index 00000000..1a9f7b33
--- /dev/null
+++ b/backend/routes/__init__.py
@@ -0,0 +1,22 @@
+"""
+路由注册模块 — 将所有 Blueprint 注册到 Flask app
+"""
+
+from .auth import bp as auth_bp
+from .upload import bp as upload_bp
+from .questions import bp as questions_bp
+from .chat import bp as chat_bp
+from .stats import bp as stats_bp
+from .settings import bp as settings_bp
+from .notes import bp as notes_bp
+
+
+def register_routes(app):
+    """注册所有 Blueprint 到 Flask app"""
+    app.register_blueprint(auth_bp, url_prefix='/api/auth')
+    app.register_blueprint(upload_bp, url_prefix='/api')
+    app.register_blueprint(questions_bp, url_prefix='/api')
+    app.register_blueprint(chat_bp, url_prefix='/api')
+    app.register_blueprint(stats_bp, url_prefix='/api')
+    app.register_blueprint(settings_bp, url_prefix='/api')
+    app.register_blueprint(notes_bp, url_prefix='/api/notes')
diff --git a/backend/routes/auth.py b/backend/routes/auth.py
new file mode 100644
index 00000000..01efeeb1
--- /dev/null
+++ b/backend/routes/auth.py
@@ -0,0 +1,284 @@
+import hmac
+import hashlib
+import logging
+import secrets
+from datetime import datetime, timedelta
+
+from flask import Blueprint, request, jsonify, session, current_app
+from werkzeug.security import generate_password_hash, check_password_hash
+
+from core.config import settings
+from core.mail import send_smtp_email
+from db import SessionLocal
+from db import crud
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint("auth", __name__)
+
+
+def _hash_registration_code(email: str, code: str, pepper: str) -> str:
+    raw = f"{email.lower().strip()}:{code}".encode("utf-8")
+    return hmac.new(pepper.encode("utf-8"), raw, hashlib.sha256).hexdigest()
+
+
+def _generate_six_digit_code() -> str:
+    return f"{secrets.randbelow(900000) + 100000}"
+
+
+@bp.route("/register", methods=["POST"])
+def auth_register():
+    """用户注册（需先调用 /send-code?type=register 并填写正确验证码）。"""
+    data = request.get_json() or {}
+    email = (data.get("email") or "").strip().lower()
+    username = (data.get("username") or "").strip()
+    password = data.get("password") or ""
+    code = (data.get("code") or "").strip()
+
+    if not email or "@" not in email:
+        return jsonify({"success": False, "error": "邮箱格式不正确"}), 400
+    if not username:
+        return jsonify({"success": False, "error": "用户名不能为空"}), 400
+    if len(password) < 6:
+        return jsonify({"success": False, "error": "密码至少 6 位"}), 400
+    if not code or not code.isdigit() or len(code) != 6:
+        return jsonify({"success": False, "error": "请输入 6 位邮箱验证码"}), 400
+
+    pepper = (current_app.secret_key or "dev-secret-change-in-production")[:64]
+    submitted_hash = _hash_registration_code(email, code, pepper)
+    now = datetime.utcnow()
+
+    user_id = None
+    username_out = None
+    email_out = None
+    is_admin_out = None
+
+    with SessionLocal() as db:
+        if crud.get_user_by_email(db, email):
+            return jsonify({"success": False, "error": "该邮箱已注册"}), 409
+
+        row = crud.get_verification_by_email(db, email)
+        if not row:
+            return jsonify({"success": False, "error": "请先获取邮箱验证码"}), 400
+        if row.expires_at < now:
+            crud.delete_verification_by_email(db, email)
+            return jsonify({"success": False, "error": "验证码已过期，请重新获取"}), 400
+        if (row.attempts or 0) >= settings.registration_max_attempts:
+            crud.delete_verification_by_email(db, email)
+            return jsonify({"success": False, "error": "验证失败次数过多，请重新获取验证码"}), 400
+
+        if not hmac.compare_digest(row.code_hash, submitted_hash):
+            crud.increment_verification_attempts(db, email)
+            return jsonify({"success": False, "error": "验证码错误"}), 400
+
+        pwd_hash = generate_password_hash(password)
+        user = crud.create_user(db, email=email, password_hash=pwd_hash, username=username)
+        crud.delete_verification_by_email(db, email)
+        # 将需要的字段提前取出为纯 Python 变量，避免脱离 Session 后访问 ORM 对象触发 DetachedInstanceError
+        user_id = user.id
+        username_out = user.username
+        email_out = user.email
+        is_admin_out = user.is_admin
+        session_version_out = user.session_version or 0
+
+    session["user_id"] = user_id
+    session["username"] = username_out
+    session["is_admin"] = bool(is_admin_out)
+    session["session_version"] = session_version_out
+    return jsonify(
+        {
+            "success": True,
+            "user": {
+                "id": user_id,
+                "email": email_out,
+                "username": username_out,
+                "is_admin": is_admin_out,
+            },
+        }
+    ), 201
+
+
+@bp.route("/send-code", methods=["POST"])
+def send_code():
+    """发送验证码（支持注册和找回密码两种场景）。
+
+    请求体：{ email, type: 'register' | 'reset' }
+    """
+    data = request.get_json() or {}
+    email = (data.get("email") or "").strip().lower()
+    code_type = (data.get("type") or "register").strip().lower()
+
+    if not email or "@" not in email:
+        return jsonify({"success": False, "error": "邮箱格式不正确"}), 400
+
+    now = datetime.utcnow()
+    # 提前生成验证码，保证频率检查与写入在同一个 DB 事务中（消除 TOCTOU 竞态）
+    code = _generate_six_digit_code()
+    pepper = (current_app.secret_key or "dev-secret-change-in-production")[:64]
+    code_hash = _hash_registration_code(email, code, pepper)
+    expires_at = now + timedelta(minutes=settings.registration_code_ttl_minutes)
+
+    with SessionLocal() as db:
+        # 频率限制与写入在同一 session，防止并发请求绕过限制
+        row = crud.get_verification_by_email(db, email)
+        if row and row.last_sent_at:
+            delta = (now - row.last_sent_at).total_seconds()
+            if delta < settings.registration_send_interval_seconds:
+                wait = int(settings.registration_send_interval_seconds - delta)
+                return jsonify(
+                    {"success": False, "error": f"发送过于频繁，请 {wait} 秒后再试"}
+                ), 429
+
+        user = crud.get_user_by_email(db, email)
+        # 防邮箱枚举：注册时邮箱已存在、重置时邮箱不存在，均返回相同成功响应但不实际发送
+        # 同时写入记录以保证频率限制行为一致，防止通过 429 差异探测邮箱
+        if (code_type == "register" and user) or (code_type == "reset" and not user):
+            crud.upsert_registration_code(db, email, "anti-enum-dummy", now + timedelta(minutes=1), now)
+            return jsonify({"success": True, "message": "验证码已发送"})
+
+        crud.upsert_registration_code(db, email, code_hash, expires_at, now)
+
+    subject_text = "找回密码" if code_type == "reset" else "注册"
+    smtp_ok = bool(settings.smtp_host and settings.smtp_from)
+    if smtp_ok:
+        try:
+            send_smtp_email(
+                host=settings.smtp_host,
+                port=settings.smtp_port,
+                user=settings.smtp_user,
+                password=settings.smtp_password,
+                mail_from=settings.smtp_from,
+                use_tls=settings.smtp_use_tls,
+                to_addr=email,
+                subject=f"【智卷系统】{subject_text}验证码",
+                body=f"您的{subject_text}验证码为：{code}，{settings.registration_code_ttl_minutes} 分钟内有效。请勿泄露给他人。",
+                async_send=True,
+            )
+        except Exception:
+            logger.exception("发送验证码邮件失败")
+            return jsonify({"success": False, "error": "邮件发送失败，请稍后重试"}), 502
+    else:
+        if current_app.debug:
+            logger.warning(
+                "[开发模式] 未配置 SMTP，验证码（仅调试用） email=%s code=%s",
+                email,
+                code,
+            )
+        else:
+            return jsonify(
+                {"success": False, "error": "服务器未配置发信，无法发送验证码"}
+            ), 503
+
+    return jsonify({"success": True, "message": "验证码已发送"})
+
+
+@bp.route("/reset-password", methods=["POST"])
+def reset_password():
+    """通过邮箱验证码重置密码。
+
+    请求体：{ email, code, password }
+    """
+    data = request.get_json() or {}
+    email = (data.get("email") or "").strip().lower()
+    code = (data.get("code") or "").strip()
+    password = data.get("password") or ""
+
+    if not email or "@" not in email:
+        return jsonify({"success": False, "error": "邮箱格式不正确"}), 400
+    if not code or not code.isdigit() or len(code) != 6:
+        return jsonify({"success": False, "error": "请输入 6 位验证码"}), 400
+    if len(password) < 6:
+        return jsonify({"success": False, "error": "密码至少 6 位"}), 400
+
+    pepper = (current_app.secret_key or "dev-secret-change-in-production")[:64]
+    submitted_hash = _hash_registration_code(email, code, pepper)
+    now = datetime.utcnow()
+
+    with SessionLocal() as db:
+        user = crud.get_user_by_email(db, email)
+        if not user:
+            # 防枚举：与验证码错误返回相同提示
+            return jsonify({"success": False, "error": "验证码错误"}), 400
+
+        row = crud.get_verification_by_email(db, email)
+        if not row:
+            return jsonify({"success": False, "error": "请先获取验证码"}), 400
+        if row.expires_at < now:
+            crud.delete_verification_by_email(db, email)
+            return jsonify({"success": False, "error": "验证码已过期，请重新获取"}), 400
+        if (row.attempts or 0) >= settings.registration_max_attempts:
+            crud.delete_verification_by_email(db, email)
+            return jsonify({"success": False, "error": "验证失败次数过多，请重新获取验证码"}), 400
+
+        if not hmac.compare_digest(row.code_hash, submitted_hash):
+            crud.increment_verification_attempts(db, email)
+            return jsonify({"success": False, "error": "验证码错误"}), 400
+
+        if check_password_hash(user.password_hash, password):
+            return jsonify({"success": False, "error": "新密码不能与原密码相同"}), 400
+
+        new_hash = generate_password_hash(password)
+        crud.update_user_password(db, email, new_hash)
+        crud.delete_verification_by_email(db, email)
+
+    return jsonify({"success": True, "message": "密码重置成功"})
+
+
+@bp.route("/login", methods=["POST"])
+def auth_login():
+    """用户登录（支持邮箱或用户名）"""
+    data = request.get_json() or {}
+    identifier = (data.get("email") or data.get("identifier") or "").strip()
+    password = data.get("password") or ""
+
+    if not identifier:
+        return jsonify({"error": "请输入邮箱或用户名"}), 400
+
+    with SessionLocal() as db:
+        user = crud.get_user_by_login(db, identifier)
+        if not user or not check_password_hash(user.password_hash, password):
+            return jsonify({"error": "账号或密码错误"}), 401
+        session["user_id"] = user.id
+        session["username"] = user.username
+        session["is_admin"] = bool(user.is_admin)
+        session["session_version"] = user.session_version or 0
+        return jsonify(
+            {
+                "user": {
+                    "id": user.id,
+                    "email": user.email,
+                    "username": user.username,
+                    "is_admin": user.is_admin,
+                }
+            }
+        )
+
+
+@bp.route("/logout", methods=["POST"])
+def auth_logout():
+    """退出登录"""
+    session.clear()
+    return jsonify({"ok": True})
+
+
+@bp.route("/me", methods=["GET"])
+def auth_me():
+    """获取当前登录用户"""
+    user_id = session.get("user_id")
+    if not user_id:
+        return jsonify({"error": "未登录", "code": "UNAUTHORIZED"}), 401
+    with SessionLocal() as db:
+        user = crud.get_user_by_id(db, user_id)
+        if not user:
+            session.clear()
+            return jsonify({"error": "用户不存在", "code": "UNAUTHORIZED"}), 401
+        return jsonify(
+            {
+                "user": {
+                    "id": user.id,
+                    "email": user.email,
+                    "username": user.username,
+                    "is_admin": user.is_admin,
+                }
+            }
+        )
diff --git a/backend/routes/chat.py b/backend/routes/chat.py
new file mode 100644
index 00000000..d7b636c7
--- /dev/null
+++ b/backend/routes/chat.py
@@ -0,0 +1,303 @@
+"""Chat-related routes (AI conversation with questions)."""
+
+import json
+import logging
+
+from flask import Blueprint, request, jsonify, session, Response
+
+from db import SessionLocal
+from db import crud
+from db.models import Question, ChatSession as ChatSessionModel, QuestionTagMapping
+from core.config import settings
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('chat', __name__)
+
+
+def _effective_user_id():
+    """管理员返回 None（不过滤），普通用户返回 user_id"""
+    if session.get('is_admin'):
+        return None
+    return session.get('user_id')
+
+
+def _serialize_chat_session(s) -> dict:
+    """将 ChatSession ORM 对象序列化为前端 JSON"""
+    return {
+        'id': s.id,
+        'title': s.title or '新对话',
+        'question_id': s.question_id,
+        'created_at': s.created_at.isoformat() if s.created_at else None,
+        'updated_at': s.updated_at.isoformat() if s.updated_at else None,
+    }
+
+
+def _serialize_question(q: Question) -> dict:
+    """将 Question ORM 对象序列化为前端 JSON 格式"""
+    subject = None
+    if q.batch:
+        subject = q.batch.subject
+    knowledge_tags = []
+    if q.tags:
+        for mapping in q.tags:
+            if mapping.tag:
+                knowledge_tags.append(mapping.tag.tag_name)
+
+    return {
+        'id': q.id,
+        'question_type': q.question_type,
+        'content_json': json.loads(q.content_json) if q.content_json else [],
+        'options_json': json.loads(q.options_json) if q.options_json else None,
+        'has_formula': q.has_formula,
+        'has_image': q.has_image,
+        'needs_correction': q.needs_correction,
+        'answer': q.answer,
+        'subject': subject,
+        'knowledge_tags': knowledge_tags,
+        'created_at': q.created_at.isoformat() if q.created_at else None,
+    }
+
+
+@bp.route('/question/<int:question_id>/chats', methods=['GET'])
+def get_question_chats(question_id):
+    """获取某道题目的所有对话会话"""
+    try:
+        with SessionLocal() as db:
+            sessions = crud.get_chat_sessions_by_question(db, question_id, user_id=_effective_user_id())
+            return jsonify({
+                'success': True,
+                'sessions': [_serialize_chat_session(s) for s in sessions],
+            })
+    except Exception as e:
+        logger.exception("获取对话列表失败")
+        return jsonify({'success': False, 'error': '获取对话列表失败'}), 500
+
+
+@bp.route('/chat', methods=['POST'])
+def create_chat():
+    """创建新对话（支持绑定题目或独立对话）"""
+    try:
+        data = request.get_json(silent=True) or {}
+        question_id = data.get('question_id')  # 可选
+        title = data.get('title', '新对话')
+        user_id = session.get('user_id')
+
+        if question_id:
+            with SessionLocal() as db:
+                uid = _effective_user_id()
+                q_query = db.query(Question).filter(Question.id == question_id)
+                if uid is not None:
+                    q_query = q_query.filter(Question.user_id == uid)
+                question = q_query.first()
+                if not question:
+                    return jsonify({'success': False, 'error': '题目不存在'}), 404
+                chat_session = crud.create_chat_session(db, question_id=question_id, user_id=user_id, title=title)
+                return jsonify({'success': True, 'session': _serialize_chat_session(chat_session)})
+        else:
+            # 独立对话
+            with SessionLocal() as db:
+                chat_session = crud.create_chat_session(db, user_id=user_id, title=title)
+                return jsonify({'success': True, 'session': _serialize_chat_session(chat_session)})
+
+    except Exception as e:
+        logger.exception("创建对话失败")
+        return jsonify({'success': False, 'error': '创建对话失败'}), 500
+
+
+@bp.route('/chat/my-sessions', methods=['GET'])
+def get_my_chat_sessions():
+    """获取当前用户的独立对话列表"""
+    try:
+        user_id = session.get('user_id')
+        page = max(1, request.args.get('page', 1, type=int))
+        page_size = min(50, max(1, request.args.get('limit', 20, type=int)))
+
+        with SessionLocal() as db:
+            sessions_list, total = crud.get_user_chat_sessions(db, user_id, page=page, page_size=page_size)
+            return jsonify({
+                'success': True,
+                'sessions': [_serialize_chat_session(s) for s in sessions_list],
+                'total': total,
+            })
+    except Exception as e:
+        logger.exception("获取对话列表失败")
+        return jsonify({'success': False, 'error': '获取对话列表失败'}), 500
+
+
+@bp.route('/chat/<int:session_id>', methods=['PATCH'])
+def update_chat_title(session_id):
+    """更新对话标题"""
+    try:
+        data = request.get_json(silent=True) or {}
+        title = data.get('title', '').strip()
+        if not title:
+            return jsonify({'success': False, 'error': '标题不能为空'}), 400
+
+        with SessionLocal() as db:
+            s = crud.update_chat_session_title(db, session_id, title, user_id=_effective_user_id())
+            if not s:
+                return jsonify({'success': False, 'error': '对话不存在'}), 404
+            return jsonify({'success': True, 'session': _serialize_chat_session(s)})
+    except Exception as e:
+        logger.exception("更新对话标题失败")
+        return jsonify({'success': False, 'error': '更新失败'}), 500
+
+
+@bp.route('/chat/<int:session_id>', methods=['DELETE'])
+def delete_chat(session_id):
+    """删除对话"""
+    try:
+        with SessionLocal() as db:
+            if not crud.delete_chat_session(db, session_id, user_id=_effective_user_id()):
+                return jsonify({'success': False, 'error': '对话不存在'}), 404
+            return jsonify({'success': True})
+    except Exception as e:
+        logger.exception("删除对话失败")
+        return jsonify({'success': False, 'error': '删除失败'}), 500
+
+
+@bp.route('/chat/sessions', methods=['GET'])
+def get_chat_sessions():
+    """分页获取所有对话会话"""
+    try:
+        page = max(1, request.args.get('page', 1, type=int))
+        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
+
+        with SessionLocal() as db:
+            sessions, total = crud.get_all_chat_sessions(db, page=page, page_size=page_size, user_id=_effective_user_id())
+            total_pages = (total + page_size - 1) // page_size
+
+            return jsonify({
+                'success': True,
+                'sessions': [_serialize_chat_session(s) for s in sessions],
+                'total': total,
+                'page': page,
+                'total_pages': total_pages,
+            })
+
+    except Exception as e:
+        logger.exception("获取对话会话列表失败")
+        return jsonify({'success': False, 'error': '获取对话会话列表失败'}), 500
+
+
+@bp.route('/chat/<int:session_id>/messages', methods=['GET'])
+def get_chat_messages(session_id):
+    """游标分页获取对话消息"""
+    try:
+        limit = min(100, max(1, request.args.get('limit', 30, type=int)))
+        before_id = request.args.get('before_id', type=int)
+
+        with SessionLocal() as db:
+            result = crud.get_chat_messages(db, session_id, limit=limit, before_id=before_id, user_id=_effective_user_id())
+            return jsonify({
+                'success': True,
+                'messages': result['messages'],
+                'hasMore': result['hasMore'],
+            })
+
+    except Exception as e:
+        logger.exception("获取对话消息失败")
+        return jsonify({'success': False, 'error': '获取对话消息失败'}), 500
+
+
+@bp.route('/chat/<int:session_id>/stream', methods=['POST'])
+def stream_chat(session_id):
+    """SSE 流式对话"""
+    from agents.teach import stream_teach
+    from sqlalchemy.orm import selectinload
+
+    try:
+        data = request.get_json(silent=True) or {}
+        message = data.get('message', '').strip()
+        model_provider = data.get('model_provider', 'openai')
+        model_name = data.get('model_name') or None
+        deep_think = data.get('deep_think', False)
+
+        # 深度思考模式：切换到 reasoner 模型
+        if deep_think and model_provider == 'openai':
+            model_name = 'deepseek-reasoner'
+
+        if not message:
+            return jsonify({'success': False, 'error': '消息不能为空'}), 400
+
+        if not settings.is_valid_provider(model_provider):
+            return jsonify({'success': False, 'error': f'不支持的模型供应商: {model_provider}'}), 400
+
+        # 从数据库加载用户的 LLM 凭据
+        user_id = session.get('user_id')
+        if user_id:
+            settings.load_providers_from_db(user_id)
+
+        with SessionLocal() as db:
+            uid = _effective_user_id()
+            cs_query = db.query(ChatSessionModel).options(
+                selectinload(ChatSessionModel.question)
+                .selectinload(Question.batch),
+                selectinload(ChatSessionModel.question)
+                .selectinload(Question.tags)
+                .selectinload(QuestionTagMapping.tag),
+            ).filter(ChatSessionModel.id == session_id)
+            if uid is not None:
+                cs_query = cs_query.filter(ChatSessionModel.user_id == uid)
+            chat_session = cs_query.first()
+            if not chat_session:
+                return jsonify({'success': False, 'error': '对话不存在'}), 404
+
+            # 独立对话或题目绑定对话
+            question = chat_session.question
+            q_data = _serialize_question(question) if question else None
+
+            # 加载历史消息（最近 20 条）
+            history_result = crud.get_chat_messages(db, session_id, limit=20)
+            history = [{"role": m["role"], "content": m["content"]} for m in history_result["messages"]]
+
+            # 追加用户消息
+            history.append({"role": "user", "content": message})
+            crud.add_chat_message(db, session_id, "user", message)
+
+        def generate():
+            full_response = []
+            full_reasoning = []
+            try:
+                for chunk in stream_teach(
+                    question=q_data,
+                    messages=history,
+                    provider=model_provider,
+                    model_name=model_name,
+                ):
+                    # stream_teach 现在 yield dict: {"type": "reasoning"|"content", "content": "..."}
+                    if isinstance(chunk, dict):
+                        if chunk["type"] == "reasoning":
+                            full_reasoning.append(chunk["content"])
+                            yield f"data: {json.dumps({'reasoning': chunk['content']}, ensure_ascii=False)}\n\n"
+                        else:
+                            full_response.append(chunk["content"])
+                            yield f"data: {json.dumps({'token': chunk['content']}, ensure_ascii=False)}\n\n"
+                    else:
+                        # 兼容旧格式（纯字符串）
+                        full_response.append(chunk)
+                        yield f"data: {json.dumps({'token': chunk}, ensure_ascii=False)}\n\n"
+            except Exception as e:
+                logger.exception("流式对话错误")
+                yield f"data: {json.dumps({'error': str(e)}, ensure_ascii=False)}\n\n"
+
+            # 保存完整的 assistant 回复（思考过程不保存到消息历史）
+            assistant_content = "".join(full_response)
+            if assistant_content:
+                try:
+                    with SessionLocal() as db:
+                        crud.add_chat_message(db, session_id, "assistant", assistant_content)
+                except Exception as e:
+                    logger.error(f"保存 assistant 回复失败: {e}")
+
+            yield f"data: {json.dumps({'done': True})}\n\n"
+
+        resp = Response(generate(), mimetype='text/event-stream')
+        resp.headers['Cache-Control'] = 'no-cache'
+        resp.headers['X-Accel-Buffering'] = 'no'
+        return resp
+
+    except Exception as e:
+        logger.exception("流式对话失败")
+        return jsonify({'success': False, 'error': '对话失败，请稍后重试'}), 500
diff --git a/backend/routes/notes.py b/backend/routes/notes.py
new file mode 100644
index 00000000..cb16d88d
--- /dev/null
+++ b/backend/routes/notes.py
@@ -0,0 +1,250 @@
+"""
+笔记模块 API 路由
+
+流程：上传笔记图片 → PaddleOCR 识别 → LLM 整理 → 保存到数据库
+"""
+
+import os
+import uuid
+import json
+import logging
+from pathlib import PurePath
+
+from flask import Blueprint, request, jsonify, session
+
+from core.config import settings
+from db import SessionLocal
+from db import crud
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('notes', __name__)
+
+
+def _effective_user_id():
+    """管理员返回 None（不过滤），普通用户返回 user_id"""
+    if session.get('is_admin'):
+        return None
+    return session.get('user_id')
+
+
+def _allowed_image(filename):
+    """检查是否为允许的图片格式"""
+    return PurePath(filename).suffix.lower().lstrip('.') in {'png', 'jpg', 'jpeg', 'bmp', 'tiff', 'webp'}
+
+
+def _serialize_note(note) -> dict:
+    """将 Note ORM 对象序列化为前端 JSON"""
+    knowledge_tags = []
+    if note.tags:
+        for mapping in note.tags:
+            if mapping.tag:
+                knowledge_tags.append(mapping.tag.tag_name)
+
+    return {
+        'id': note.id,
+        'title': note.title,
+        'subject': note.subject,
+        'content_markdown': note.content_markdown,
+        'source_images': json.loads(note.source_images_json) if note.source_images_json else [],
+        'knowledge_tags': knowledge_tags,
+        'created_at': note.created_at.isoformat() if note.created_at else None,
+        'updated_at': note.updated_at.isoformat() if note.updated_at else None,
+    }
+
+
+@bp.route('/', methods=['POST'])
+def create_note():
+    """上传笔记图片 → OCR → LLM 整理 → 保存
+
+    请求：multipart/form-data
+      - files: 一张或多张图片
+      - model_provider: 模型供应商（可选，默认 openai）
+      - model_name: 模型名称（可选）
+    """
+    files = request.files.getlist('files')
+    if not files or all(f.filename == '' for f in files):
+        return jsonify({'success': False, 'error': '请上传笔记图片'}), 400
+
+    # 校验文件格式
+    for f in files:
+        if f.filename and not _allowed_image(f.filename):
+            return jsonify({'success': False, 'error': f'不支持的图片格式: {f.filename}'}), 400
+
+    user_id = session.get('user_id')
+    model_provider = request.form.get('model_provider', 'openai')
+    model_name = request.form.get('model_name') or None
+
+    try:
+        # 1. 保存上传图片
+        saved_paths = []
+        for f in files:
+            if not f.filename:
+                continue
+            ext = os.path.splitext(f.filename)[1].lower()
+            filename = f"{uuid.uuid4().hex}{ext}"
+            filepath = os.path.join(str(settings.upload_dir), filename)
+            f.save(filepath)
+            saved_paths.append(filepath)
+
+        if not saved_paths:
+            return jsonify({'success': False, 'error': '没有有效的图片文件'}), 400
+
+        # 2. OCR 识别
+        from src.paddleocr_client import PaddleOCRClient
+        from src.utils import simplify_ocr_results, run_async
+
+        # 从数据库加载用户 OCR 凭据
+        ocr_kwargs = {}
+        if user_id:
+            with SessionLocal() as db:
+                ocr_provider = crud.get_active_provider(db, user_id, 'paddleocr')
+                if ocr_provider:
+                    ocr_kwargs = {
+                        "api_url": ocr_provider.base_url,
+                        "token": ocr_provider.api_key,
+                        "model": ocr_provider.model_name,
+                    }
+
+        client = PaddleOCRClient(**ocr_kwargs)
+        raw_results = run_async(client.parse_images_async(saved_paths))
+        simplified = simplify_ocr_results(raw_results)
+
+        # 拼接 OCR 文本 + 图片引用
+        ocr_text_parts = []
+        image_refs = []
+        for page in simplified:
+            for block in page.get('blocks', []):
+                label = block.get('block_label', '')
+                content = block.get('block_content', '').strip()
+                if label in ('image', 'chart') and content:
+                    # 图片块：记录路径，在文本中插入占位符
+                    image_refs.append(content)
+                    ocr_text_parts.append(f'[图片: {content}]')
+                elif content:
+                    ocr_text_parts.append(content)
+        ocr_text = '\n'.join(ocr_text_parts)
+
+        if not ocr_text.strip():
+            return jsonify({'success': False, 'error': 'OCR 未识别到任何文字，请检查图片'}), 400
+
+        # 3. LLM 整理
+        if user_id:
+            settings.load_providers_from_db(user_id)
+
+        from agents.note import invoke_note_organize
+        result = invoke_note_organize(ocr_text, provider=model_provider, model_name=model_name)
+
+        # 4. 保存到数据库
+        with SessionLocal() as db:
+            note = crud.save_note(
+                db,
+                title=result.title,
+                subject=result.subject,
+                content_markdown=result.content_markdown,
+                source_images=saved_paths,
+                ocr_text=ocr_text,
+                knowledge_tags=result.knowledge_tags,
+                user_id=user_id,
+            )
+            return jsonify({
+                'success': True,
+                'note': _serialize_note(note),
+            }), 201
+
+    except Exception as e:
+        logger.exception("笔记创建失败")
+        return jsonify({'success': False, 'error': f'笔记创建失败: {str(e)}'}), 500
+
+
+@bp.route('/', methods=['GET'])
+def list_notes():
+    """分页查询笔记列表
+
+    查询参数：page, limit, subject, knowledge_tag, keyword
+    """
+    try:
+        page = max(1, request.args.get('page', 1, type=int))
+        limit = min(100, max(1, request.args.get('limit', 20, type=int)))
+        subject = request.args.get('subject') or None
+        knowledge_tag = request.args.get('knowledge_tag') or None
+        keyword = request.args.get('keyword') or None
+        user_id = session.get('user_id')
+
+        with SessionLocal() as db:
+            notes, total = crud.get_notes(
+                db,
+                user_id=user_id,
+                subject=subject,
+                knowledge_tag=knowledge_tag,
+                keyword=keyword,
+                page=page,
+                page_size=limit,
+            )
+            return jsonify({
+                'success': True,
+                'items': [_serialize_note(n) for n in notes],
+                'total': total,
+                'page': page,
+                'total_pages': (total + limit - 1) // limit,
+            })
+
+    except Exception as e:
+        logger.exception("获取笔记列表失败")
+        return jsonify({'success': False, 'error': '获取笔记列表失败'}), 500
+
+
+@bp.route('/<int:note_id>', methods=['GET'])
+def get_note(note_id):
+    """获取单条笔记详情"""
+    try:
+        with SessionLocal() as db:
+            note = crud.get_note_by_id(db, note_id, user_id=_effective_user_id())
+            if not note:
+                return jsonify({'success': False, 'error': '笔记不存在'}), 404
+            return jsonify({
+                'success': True,
+                'note': _serialize_note(note),
+            })
+    except Exception as e:
+        logger.exception("获取笔记详情失败")
+        return jsonify({'success': False, 'error': '获取笔记详情失败'}), 500
+
+
+@bp.route('/<int:note_id>', methods=['PATCH'])
+def update_note_route(note_id):
+    """更新笔记（标题、内容、科目、标签）"""
+    try:
+        data = request.get_json(silent=True) or {}
+        with SessionLocal() as db:
+            note = crud.update_note(
+                db,
+                note_id=note_id,
+                title=data.get('title'),
+                content_markdown=data.get('content_markdown'),
+                subject=data.get('subject'),
+                knowledge_tags=data.get('knowledge_tags'),
+                user_id=_effective_user_id(),
+            )
+            if not note:
+                return jsonify({'success': False, 'error': '笔记不存在'}), 404
+            return jsonify({
+                'success': True,
+                'note': _serialize_note(note),
+            })
+    except Exception as e:
+        logger.exception("更新笔记失败")
+        return jsonify({'success': False, 'error': '更新笔记失败'}), 500
+
+
+@bp.route('/<int:note_id>', methods=['DELETE'])
+def delete_note_route(note_id):
+    """删除笔记"""
+    try:
+        with SessionLocal() as db:
+            if not crud.delete_note(db, note_id, user_id=_effective_user_id()):
+                return jsonify({'success': False, 'error': '笔记不存在'}), 404
+            return jsonify({'success': True})
+    except Exception as e:
+        logger.exception("删除笔记失败")
+        return jsonify({'success': False, 'error': '删除笔记失败'}), 500
diff --git a/backend/routes/questions.py b/backend/routes/questions.py
new file mode 100644
index 00000000..2f0e8ece
--- /dev/null
+++ b/backend/routes/questions.py
@@ -0,0 +1,660 @@
+import os
+import json
+import logging
+from datetime import datetime
+
+from flask import Blueprint, request, jsonify, session
+
+from db import SessionLocal
+from db import crud
+from db.models import Question, UploadBatch, KnowledgeTag
+from core.config import settings
+from src.utils import export_wrongbook as export_wrongbook_md
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('questions', __name__)
+
+
+def _effective_user_id():
+    """管理员返回 None（不过滤），普通用户返回 user_id"""
+    if session.get('is_admin'):
+        return None
+    return session.get('user_id')
+
+
+def _serialize_question(q: Question) -> dict:
+    """将 Question ORM 对象序列化为前端 JSON 格式"""
+    subject = None
+    if q.batch:
+        subject = q.batch.subject
+    knowledge_tags = []
+    if q.tags:
+        for mapping in q.tags:
+            if mapping.tag:
+                knowledge_tags.append(mapping.tag.tag_name)
+
+    return {
+        'id': q.id,
+        'question_type': q.question_type,
+        'content_json': json.loads(q.content_json) if q.content_json else [],
+        'options_json': json.loads(q.options_json) if q.options_json else None,
+        'has_formula': q.has_formula,
+        'has_image': q.has_image,
+        'needs_correction': q.needs_correction,
+        'answer': q.answer,
+        'subject': subject,
+        'knowledge_tags': knowledge_tags,
+        'created_at': q.created_at.isoformat() if q.created_at else None,
+    }
+
+
+def _serialize_question_detail(q: Question) -> dict:
+    """将 Question ORM 对象序列化为详情 JSON（含科目、标签、答案等）"""
+    base = _serialize_question(q)
+    # subject / knowledge_tags already set by _serialize_question
+    base['original_filename'] = q.batch.original_filename if q.batch else None
+    base['user_answer'] = q.user_answer
+    base['updated_at'] = q.updated_at.isoformat() if q.updated_at else None
+    base['review_status'] = q.review_status or '待复习'
+    base['image_refs_json'] = json.loads(q.image_refs_json) if q.image_refs_json else None
+    return base
+
+
+def _read_split_subject():
+    """从 split_metadata.json 读取学科信息"""
+    meta_path = os.path.join(str(settings.results_dir), "split_metadata.json")
+    if os.path.exists(meta_path):
+        with open(meta_path, 'r', encoding='utf-8') as f:
+            return json.load(f).get("subject")
+    return None
+
+
+@bp.route('/questions', methods=['GET'])
+def get_questions():
+    """
+    获取已分割的题目列表
+
+    Returns:
+        JSON响应，包含题目列表
+    """
+    try:
+        results_dir = settings.results_dir
+        questions_file = os.path.join(results_dir, "questions.json")
+
+        if not os.path.exists(questions_file):
+            return jsonify({
+                'success': True,
+                'questions': [],
+                'message': '暂无题目数据'
+            })
+
+        with open(questions_file, 'r', encoding='utf-8') as f:
+            questions = json.load(f)
+
+        return jsonify({
+            'success': True,
+            'questions': questions
+        })
+
+    except json.JSONDecodeError:
+        return jsonify({
+            'success': False,
+            'error': '题目数据文件格式错误，请重新分割题目'
+        }), 500
+    except Exception as e:
+        logger.exception("获取题目列表失败")
+        return jsonify({
+            'success': False,
+            'error': '获取题目列表失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/history', methods=['GET'])
+def get_history():
+    """
+    分页查询历史题目（全部题目，非仅错题）
+
+    Query参数:
+        page: 页码（默认1）
+        page_size: 每页数量（默认20）
+        start_date: 开始日期（可选，格式：YYYY-MM-DD）
+        end_date: 结束日期（可选，格式：YYYY-MM-DD）
+    """
+    try:
+        page = max(1, request.args.get('page', 1, type=int))
+        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
+        start_date_str = request.args.get('start_date')
+        end_date_str = request.args.get('end_date')
+
+        # 解析日期
+        start_date = None
+        end_date = None
+        if start_date_str:
+            start_date = datetime.strptime(start_date_str, '%Y-%m-%d')
+        if end_date_str:
+            end_date = datetime.strptime(end_date_str, '%Y-%m-%d')
+
+        with SessionLocal() as db:
+            questions, total = crud.get_history_questions(
+                db,
+                start_date=start_date,
+                end_date=end_date,
+                page=page,
+                page_size=page_size,
+                user_id=_effective_user_id(),
+            )
+
+            total_pages = (total + page_size - 1) // page_size
+            items = [_serialize_question(q) for q in questions]
+
+            return jsonify({
+                'success': True,
+                'items': items,
+                'total': total,
+                'page': page,
+                'page_size': page_size,
+                'total_pages': total_pages
+            })
+
+    except ValueError:
+        return jsonify({'success': False, 'error': '日期格式错误，请使用 YYYY-MM-DD 格式'}), 400
+    except Exception as e:
+        logger.exception("查询历史题目失败")
+        return jsonify({'success': False, 'error': '查询失败，请稍后重试'}), 500
+
+
+@bp.route('/search', methods=['GET'])
+def search_questions():
+    """
+    搜索题目（知识点/题型/关键字）
+
+    Query参数:
+        keyword: 关键字搜索（匹配题目内容）
+        knowledge_tag: 知识点标签筛选
+        question_type: 题型筛选
+        page: 页码（默认1）
+        page_size: 每页数量（默认20）
+    """
+    try:
+        page = max(1, request.args.get('page', 1, type=int))
+        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
+        keyword = request.args.get('keyword', type=str)
+        knowledge_tag = request.args.get('knowledge_tag', type=str)
+        question_type = request.args.get('question_type', type=str)
+
+        # 至少需要一个搜索条件
+        if not any([keyword, knowledge_tag, question_type]):
+            return jsonify({
+                'success': False,
+                'error': '至少需要提供一个搜索条件（keyword/knowledge_tag/question_type）'
+            }), 400
+
+        with SessionLocal() as db:
+            questions, total = crud.search_questions(
+                db,
+                keyword=keyword if keyword else None,
+                knowledge_tag=knowledge_tag if knowledge_tag else None,
+                question_type=question_type if question_type else None,
+                page=page,
+                page_size=page_size,
+                user_id=_effective_user_id(),
+            )
+
+            total_pages = (total + page_size - 1) // page_size
+            items = [_serialize_question(q) for q in questions]
+
+            return jsonify({
+                'success': True,
+                'items': items,
+                'total': total,
+                'page': page,
+                'page_size': page_size,
+                'total_pages': total_pages
+            })
+
+    except Exception as e:
+        logger.exception("搜索题目失败")
+        return jsonify({'success': False, 'error': '搜索失败，请稍后重试'}), 500
+
+
+@bp.route('/question/<int:question_id>', methods=['DELETE'])
+def delete_question(question_id):
+    """
+    删除题目
+
+    Path参数:
+        question_id: 题目ID
+    """
+    try:
+        with SessionLocal() as db:
+            success = crud.delete_question(db, question_id, user_id=_effective_user_id())
+            if success:
+                return jsonify({
+                    'success': True,
+                    'message': '题目已删除'
+                })
+            else:
+                return jsonify({
+                    'success': False,
+                    'error': '题目不存在'
+                }), 404
+
+    except Exception as e:
+        logger.exception("删除题目失败")
+        return jsonify({'success': False, 'error': '删除失败，请稍后重试'}), 500
+
+
+@bp.route('/error-bank', methods=['GET'])
+def get_error_bank():
+    """
+    错题库统一查询
+
+    Query参数:
+        subject: 科目筛选
+        knowledge_tag: 知识点标签筛选
+        question_type: 题型筛选
+        keyword: 关键字搜索
+        start_date: 开始日期（YYYY-MM-DD）
+        end_date: 结束日期（YYYY-MM-DD）
+        page: 页码（默认1）
+        page_size: 每页数量（默认20）
+    """
+    try:
+        page = max(1, request.args.get('page', 1, type=int))
+        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
+        subject = request.args.get('subject', type=str) or None
+        knowledge_tag = request.args.get('knowledge_tag', type=str) or None
+        question_type = request.args.get('question_type', type=str) or None
+        keyword = request.args.get('keyword', type=str) or None
+        review_status = request.args.get('review_status', type=str) or None
+        start_date_str = request.args.get('start_date')
+        end_date_str = request.args.get('end_date')
+
+        start_date = None
+        end_date = None
+        if start_date_str:
+            start_date = datetime.strptime(start_date_str, '%Y-%m-%d')
+        if end_date_str:
+            end_date = datetime.strptime(end_date_str, '%Y-%m-%d')
+
+        with SessionLocal() as db:
+            questions, total, grand_total = crud.query_questions(
+            db,
+            user_id=_effective_user_id(),
+                subject=subject,
+                knowledge_tag=knowledge_tag,
+                question_type=question_type,
+                keyword=keyword,
+                start_date=start_date,
+                end_date=end_date,
+                review_status=review_status,
+                page=page,
+                page_size=page_size,
+            )
+
+            total_pages = (total + page_size - 1) // page_size
+            items = [_serialize_question_detail(q) for q in questions]
+
+            return jsonify({
+                'success': True,
+                'items': items,
+                'total': total,
+                'grand_total': grand_total,
+                'page': page,
+                'page_size': page_size,
+                'total_pages': total_pages,
+            })
+
+    except ValueError:
+        return jsonify({'success': False, 'error': '日期格式错误，请使用 YYYY-MM-DD 格式'}), 400
+    except Exception as e:
+        logger.exception("查询错题库失败")
+        return jsonify({'success': False, 'error': '查询失败，请稍后重试'}), 500
+
+
+@bp.route('/subjects', methods=['GET'])
+def get_subjects():
+    """获取所有科目列表"""
+    try:
+        with SessionLocal() as db:
+            subjects = crud.get_existing_subjects(db, user_id=_effective_user_id())
+            return jsonify({'success': True, 'subjects': subjects})
+    except Exception as e:
+        logger.exception("获取科目列表失败")
+        return jsonify({'success': False, 'error': '获取科目列表失败'}), 500
+
+
+@bp.route('/question-types', methods=['GET'])
+def get_question_types():
+    """获取所有题型列表"""
+    try:
+        with SessionLocal() as db:
+            types = crud.get_existing_question_types(db, user_id=_effective_user_id())
+            return jsonify({'success': True, 'question_types': types})
+    except Exception as e:
+        logger.exception("获取题型列表失败")
+        return jsonify({'success': False, 'error': '获取题型列表失败'}), 500
+
+
+@bp.route('/question/<int:question_id>', methods=['PATCH'])
+def update_question(question_id):
+    """编辑题目内容和/或答案"""
+    try:
+        data = request.get_json(silent=True) or {}
+        with SessionLocal() as db:
+            question = db.get(Question, question_id)
+            if not question:
+                return jsonify({'success': False, 'error': '题目不存在'}), 404
+            uid = _effective_user_id()
+            if uid is not None and question.user_id != uid:
+                return jsonify({'success': False, 'error': '题目不存在'}), 404
+            try:
+                if 'content' in data:
+                    text = str(data['content'])[:20000]
+                    blocks = [{'block_type': 'text', 'content': text}]
+                    question.content_json = json.dumps(blocks, ensure_ascii=False)
+                    question.content_hash = crud.compute_content_hash(blocks)
+                if 'answer' in data:
+                    question.answer = str(data['answer'])[:10000] if data['answer'] else None
+                question.updated_at = datetime.utcnow()
+                db.commit()
+                return jsonify({'success': True})
+            except Exception:
+                db.rollback()
+                raise
+    except Exception:
+        logger.exception("编辑题目失败")
+        return jsonify({'success': False, 'error': '保存失败，请稍后重试'}), 500
+
+
+@bp.route('/question/<int:question_id>/answer', methods=['PATCH'])
+def update_question_answer(question_id):
+    """保存用户答案"""
+    try:
+        data = request.get_json(silent=True) or {}
+        user_answer = data.get('user_answer')
+        if user_answer is None:
+            return jsonify({'success': False, 'error': '缺少 user_answer 字段'}), 400
+        if len(user_answer) > 10000:
+            return jsonify({'success': False, 'error': '答案内容过长（最多 10000 字符）'}), 400
+
+        with SessionLocal() as db:
+            question = crud.update_user_answer(db, question_id, user_answer, user_id=_effective_user_id())
+            if not question:
+                return jsonify({'success': False, 'error': '题目不存在'}), 404
+
+            return jsonify({
+                'success': True,
+                'message': '答案已保存',
+                'user_answer': question.user_answer,
+                'updated_at': question.updated_at.isoformat() if question.updated_at else None,
+            })
+
+    except Exception as e:
+        logger.exception("保存答案失败")
+        return jsonify({'success': False, 'error': '保存答案失败，请稍后重试'}), 500
+
+
+@bp.route('/question/<int:question_id>/review-status', methods=['PATCH'])
+def update_question_review_status(question_id):
+    """更新题目复习状态"""
+    try:
+        data = request.get_json(silent=True) or {}
+        review_status = data.get('review_status')
+        if not review_status:
+            return jsonify({'success': False, 'error': '缺少 review_status 字段'}), 400
+
+        with SessionLocal() as db:
+            question = crud.update_review_status(db, question_id, review_status, user_id=_effective_user_id())
+            if not question:
+                return jsonify({'success': False, 'error': '题目不存在'}), 404
+
+            return jsonify({
+                'success': True,
+                'message': '复习状态已更新',
+                'review_status': question.review_status,
+                'updated_at': question.updated_at.isoformat() if question.updated_at else None,
+            })
+
+    except ValueError:
+        return jsonify({'success': False, 'error': f'无效的复习状态，可选值: {", ".join(crud.VALID_REVIEW_STATUSES)}'}), 400
+    except Exception as e:
+        logger.exception("更新复习状态失败")
+        return jsonify({'success': False, 'error': '更新复习状态失败，请稍后重试'}), 500
+
+
+@bp.route('/save-to-db', methods=['POST'])
+def save_to_db():
+    """
+    将分割好的题目导入错题库（仅入库，不导出 Markdown）
+
+    Request Body:
+        {
+            "selected_ids": ["q_0", "q_1", ...]   # 选中的题目 ID 列表
+        }
+    """
+    try:
+        data = request.get_json(silent=True) or {}
+        selected_uids = data.get('selected_ids', [])
+
+        if not isinstance(selected_uids, list) or not selected_uids:
+            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
+
+        results_dir = settings.results_dir
+        questions_file = os.path.join(results_dir, "questions.json")
+        if not os.path.exists(questions_file):
+            return jsonify({'success': False, 'error': '请先分割题目'}), 400
+
+        with open(questions_file, 'r', encoding='utf-8') as f:
+            questions = json.load(f)
+
+        uid_set = set(selected_uids)
+        selected_questions = [q for q in questions if q.get('uid') in uid_set]
+        if not selected_questions:
+            return jsonify({'success': False, 'error': '未找到选中的题目'}), 400
+
+        # 合并前端传来的 answer/user_answer（按 uid 匹配）
+        answers_map = {a['uid']: a for a in data.get('answers', []) if 'uid' in a}
+        for sq in selected_questions:
+            uid = sq.get('uid')
+            if uid and uid in answers_map:
+                if 'answer' in answers_map[uid]:
+                    sq['answer'] = answers_map[uid]['answer']
+                if 'user_answer' in answers_map[uid]:
+                    sq['user_answer'] = answers_map[uid]['user_answer']
+
+        # 读取科目信息
+        subject = _read_split_subject()
+
+        from core.state import session_lock, get_user_session
+        uid = session.get('user_id')
+        with session_lock:
+            us = get_user_session(uid)
+            batch_info = {
+                "original_filename": ", ".join(
+                    us["session_files"].get(k, {}).get("filename", "未知")
+                    for k in us["session_file_order"]
+                ),
+                "subject": subject,
+                "file_path": "",
+            }
+
+        with SessionLocal() as db:
+            result = crud.save_questions_to_db(db, selected_questions, batch_info, user_id=session.get('user_id'))
+
+        return jsonify({
+            'success': True,
+            'message': f'已导入 {result["created"]} 道题目（跳过 {result["duplicates"]} 道重复）',
+            'created': result['created'],
+            'duplicates': result['duplicates'],
+        })
+
+    except Exception as e:
+        logger.exception("导入错题库失败")
+        return jsonify({'success': False, 'error': '导入错题库失败，请稍后重试'}), 500
+
+
+@bp.route('/ai-analysis', methods=['POST'])
+def ai_analysis():
+    """
+    AI 错题分析（骨架路由）
+
+    接收一组题目 ID，调用 AI 对这些错题进行综合分析，返回：
+    - 薄弱知识点诊断
+    - 错因归纳
+    - 针对性复习建议
+
+    Request Body:
+        {
+            "question_ids": [1, 2, 3]   # 必填，至少 1 道，最多 20 道
+        }
+
+    Response:
+        {
+            "success": true,
+            "analysis": {
+                "summary": "综合诊断摘要",
+                "weak_points": ["知识点A", "知识点B"],
+                "error_patterns": [
+                    {"pattern": "错因类型", "description": "详细描述", "question_ids": [1, 2]}
+                ],
+                "suggestions": ["建议1", "建议2"],
+                "per_question": [
+                    {"question_id": 1, "diagnosis": "该题错因分析", "hint": "解题提示"}
+                ]
+            }
+        }
+
+    TODO（后端开发者）:
+        1. 从数据库查询题目详情（content_json, options_json, user_answer, knowledge_tags）
+        2. 构建 prompt，将题目内容 + 用户答案 + 知识点标签传给 LLM
+        3. 调用 LLM（建议使用 llm.py 中的 init_model，支持 openai/anthropic）
+        4. 解析 LLM 返回的结构化结果，填充 analysis 字段
+        5. 可选：将分析结果缓存到数据库，避免重复分析
+    """
+    try:
+        data = request.get_json(silent=True) or {}
+        question_ids = data.get('question_ids', [])
+
+        if not isinstance(question_ids, list) or not question_ids:
+            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
+
+        if len(question_ids) > 20:
+            return jsonify({'success': False, 'error': '单次最多分析 20 道题目'}), 400
+
+        with SessionLocal() as db:
+            questions = crud.get_questions_by_ids(db, question_ids, user_id=_effective_user_id())
+            if not questions:
+                return jsonify({'success': False, 'error': '未找到对应题目'}), 404
+
+            # ── TODO: 替换为真实 AI 分析逻辑 ──────────────────
+            # 当前返回占位数据，供前端联调使用
+            knowledge_tags = set()
+            per_question = []
+            for q in questions:
+                tags = [m.tag.tag_name for m in (q.tags or []) if m.tag]
+                knowledge_tags.update(tags)
+                per_question.append({
+                    'question_id': q.id,
+                    'diagnosis': f'题目 #{q.id} 的错因分析（待 AI 生成）',
+                    'hint': f'题目 #{q.id} 的解题提示（待 AI 生成）',
+                })
+
+            analysis = {
+                'summary': f'已选择 {len(questions)} 道题目进行分析（AI 分析功能待接入）',
+                'weak_points': list(knowledge_tags) or ['暂无知识点数据'],
+                'error_patterns': [
+                    {
+                        'pattern': '待 AI 识别',
+                        'description': '错因模式将由 AI 自动归纳',
+                        'question_ids': [q.id for q in questions],
+                    }
+                ],
+                'suggestions': [
+                    '建议回顾相关知识点章节',
+                    '建议针对薄弱环节进行专项练习',
+                ],
+                'per_question': per_question,
+            }
+            # ── END TODO ──────────────────────────────────────
+
+            return jsonify({
+                'success': True,
+                'analysis': analysis,
+            })
+
+    except Exception as e:
+        logger.exception("AI 分析失败")
+        return jsonify({'success': False, 'error': 'AI 分析失败，请稍后重试'}), 500
+
+
+@bp.route('/export-from-db', methods=['POST'])
+def export_from_db():
+    """从数据库按 ID 列表导出 Markdown 错题本"""
+    try:
+        data = request.get_json(silent=True) or {}
+        selected_ids = data.get('selected_ids', [])
+
+        if not isinstance(selected_ids, list) or not selected_ids:
+            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
+
+        with SessionLocal() as db:
+            questions_orm = crud.get_questions_by_ids(db, selected_ids, user_id=_effective_user_id())
+            if not questions_orm:
+                return jsonify({'success': False, 'error': '未找到对应题目'}), 404
+
+            # 转换为 export_wrongbook_md 所需的 dict 格式
+            questions = []
+            for q in questions_orm:
+                questions.append({
+                    'question_id': str(q.id),
+                    'question_type': q.question_type,
+                    'content_blocks': json.loads(q.content_json) if q.content_json else [],
+                    'options': json.loads(q.options_json) if q.options_json else None,
+                    'has_formula': q.has_formula,
+                    'has_image': q.has_image,
+                    'knowledge_tags': [m.tag.tag_name for m in (q.tags or []) if m.tag],
+                })
+
+            # DB 导出的题目已按 selected_ids 过滤完毕，赋顺序 uid 后全部导出
+            for i, q in enumerate(questions):
+                q['uid'] = str(i)
+            output_path = export_wrongbook_md(questions, [str(i) for i in range(len(questions))])
+
+        return jsonify({
+            'success': True,
+            'message': '错题本导出成功',
+            'output_path': output_path,
+        })
+
+    except Exception as e:
+        logger.exception("从数据库导出失败")
+        return jsonify({'success': False, 'error': '导出失败，请稍后重试'}), 500
+
+
+@bp.route('/question/<int:question_id>/answer', methods=['PUT'])
+def save_question_answer(question_id):
+    """保存题目答案（Markdown 格式，用于 AI 辅导）"""
+    try:
+        data = request.get_json(silent=True) or {}
+        answer = data.get('answer')
+        if answer is None:
+            return jsonify({'success': False, 'error': '缺少 answer 字段'}), 400
+        if len(answer) > 50000:
+            return jsonify({'success': False, 'error': '答案内容过长（最多 50000 字符）'}), 400
+
+        with SessionLocal() as db:
+            question = crud.update_question_answer(db, question_id, answer, user_id=_effective_user_id())
+            if not question:
+                return jsonify({'success': False, 'error': '题目不存在'}), 404
+
+            return jsonify({
+                'success': True,
+                'message': '答案已保存',
+                'answer': question.answer,
+            })
+
+    except Exception as e:
+        logger.exception("保存答案失败")
+        return jsonify({'success': False, 'error': '保存答案失败，请稍后重试'}), 500
diff --git a/backend/routes/settings.py b/backend/routes/settings.py
new file mode 100644
index 00000000..163a9efd
--- /dev/null
+++ b/backend/routes/settings.py
@@ -0,0 +1,310 @@
+import os
+import json
+import uuid
+import logging
+
+import requests as http_requests
+from flask import Blueprint, request, jsonify, session
+
+from core.config import settings
+from db import SessionLocal
+from db import crud
+from db.models import ProviderConfig
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('settings', __name__)
+
+
+@bp.route('/models/erase', methods=['POST'])
+def erase_text():
+    """文字擦除接口
+
+    使用 GAN 模型将试卷图片中的手写笔迹擦除，还原干净的题目底图。
+
+    Request:
+        multipart/form-data，字段 file：图片文件（JPEG/PNG/BMP 等）
+
+    Response:
+        {"success": true, "result_url": "/erased/<filename>"}
+    """
+    if 'file' not in request.files:
+        return jsonify({'success': False, 'error': '缺少 file 字段'}), 400
+
+    file = request.files['file']
+    if not file or not file.filename:
+        return jsonify({'success': False, 'error': '文件为空'}), 400
+
+    ext = os.path.splitext(file.filename)[1].lower() or '.png'
+    allowed = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.webp'}
+    if ext not in allowed:
+        return jsonify({'success': False, 'error': f'不支持的图片格式: {ext}'}), 400
+
+    try:
+        image_bytes = file.read()
+    except Exception:
+        return jsonify({'success': False, 'error': '读取文件失败'}), 400
+
+    try:
+        from models.inference import InferenceEngine
+
+        result_img = InferenceEngine().run(image_bytes)
+
+        filename = uuid.uuid4().hex + '.png'
+        save_path = settings.erased_dir / filename
+        result_img.save(save_path, format='PNG')
+
+        return jsonify({'success': True, 'result_url': f'/erased/{filename}'})
+
+    except FileNotFoundError as e:
+        logger.error("模型权重缺失: %s", e)
+        return jsonify({'success': False, 'error': str(e)}), 503
+    except Exception:
+        logger.exception("文字擦除失败")
+        return jsonify({'success': False, 'error': '擦除处理失败，请稍后重试'}), 500
+
+
+@bp.route('/status', methods=['GET'])
+def get_status():
+    """
+    获取系统状态
+
+    Returns:
+        JSON响应，包含系统配置和状态
+    """
+    try:
+        user_id = session.get('user_id')
+
+        # 检查 EnsExam 模型权重是否存在
+        ensexam_configured = settings.model_path.exists()
+
+        if user_id:
+            # 已登录：从数据库读取用户配置
+            with SessionLocal() as db:
+                paddle_provider = crud.get_active_provider(db, user_id, 'paddleocr')
+                paddleocr_configured = bool(paddle_provider and paddle_provider.api_key and paddle_provider.base_url)
+
+                # 构建用户级可用模型列表
+                available_models = []
+                for category, label in [('openai', 'OpenAI'), ('anthropic', 'Anthropic')]:
+                    provider = crud.get_active_provider(db, user_id, category)
+                    if provider and provider.api_key:
+                        available_models.append({
+                            'value': category,
+                            'label': provider.name or label,
+                            'configured': True,
+                            'status': '配置成功',
+                            'default_model': provider.model_name or '',
+                            'models': [provider.model_name] if provider.model_name else [],
+                        })
+                    else:
+                        available_models.append({
+                            'value': category,
+                            'label': label,
+                            'configured': False,
+                            'status': '未配置',
+                            'default_model': '',
+                            'models': [],
+                        })
+        else:
+            # 未登录：返回空状态
+            paddleocr_configured = False
+            available_models = []
+
+        status = {
+            'paddleocr_configured': paddleocr_configured,
+            'ensexam_configured': ensexam_configured,
+            'available_models': available_models,
+            'langsmith_enabled': os.getenv('LANGSMITH_TRACING', 'false').lower() == 'true',
+            'output_dirs': {
+                'pages': str(settings.pages_dir),
+                'struct': str(settings.struct_dir),
+                'results': str(settings.results_dir),
+            }
+        }
+
+        return jsonify({
+            'success': True,
+            'status': status
+        })
+
+    except Exception as e:
+        logger.exception("获取系统状态失败")
+        return jsonify({
+            'success': False,
+            'error': '获取系统状态失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/config', methods=['GET'])
+def get_config():
+    """获取当前用户的 provider 配置"""
+    try:
+        user_id = session.get('user_id')
+        if not user_id:
+            return jsonify({'success': False, 'error': '请先登录'}), 401
+        with SessionLocal() as db:
+            config = crud.get_user_providers(db, user_id)
+        return jsonify({'success': True, 'config': config})
+    except Exception as e:
+        logger.exception("获取配置失败")
+        return jsonify({'success': False, 'error': '获取配置失败'}), 500
+
+
+@bp.route('/config', methods=['PUT'])
+def update_config():
+    """保存当前用户的 provider 配置到数据库"""
+    try:
+        data = request.get_json(force=True) or {}
+        user_id = session.get('user_id')
+
+        if not user_id:
+            return jsonify({'success': False, 'error': '请先登录'}), 401
+
+        with SessionLocal() as db:
+            try:
+                crud.save_user_providers(db, user_id, data)
+            except Exception:
+                db.rollback()
+                raise
+        return jsonify({'success': True})
+
+    except Exception as e:
+        logger.exception("更新配置失败")
+        return jsonify({'success': False, 'error': '更新配置失败，请检查日志'}), 500
+
+
+@bp.route('/models/list', methods=['POST'])
+def list_models():
+    """代理请求目标 API 的模型列表，绕过浏览器 CORS 限制。
+
+    请求体:
+        type: 'openai' | 'anthropic'
+        api_key: API Key（可选，留空则使用系统已配置的 key）
+        base_url: Base URL（可选）
+        provider_id: 已有 provider 的 id（可选，用于回退读取已存 key）
+    """
+    try:
+        data = request.get_json(force=True) or {}
+        provider_type = data.get('type', 'openai')
+        api_key = data.get('api_key') or ''
+        base_url = data.get('base_url') or ''
+
+        # 如果前端没传 key，从数据库读取已保存的配置
+        if not api_key:
+            user_id = session.get('user_id')
+            provider_id = data.get('provider_id')
+            if user_id:
+                with SessionLocal() as db:
+                    if provider_id:
+                        provider = db.query(ProviderConfig).filter_by(
+                            id=provider_id, user_id=user_id
+                        ).first()
+                    else:
+                        provider = crud.get_active_provider(db, user_id, provider_type)
+                    if provider:
+                        if not api_key:
+                            api_key = provider.api_key or ''
+                        if not base_url:
+                            base_url = provider.base_url or ''
+
+        if not api_key:
+            return jsonify({'error': '未提供 API Key，请先在设置中配置'}), 400
+
+        if provider_type == 'openai':
+            url = (base_url.rstrip('/') if base_url else 'https://api.openai.com') + '/v1/models'
+            headers = {'Authorization': f'Bearer {api_key}'}
+            resp = http_requests.get(url, headers=headers, timeout=15)
+
+            # 部分 OpenAI 兼容 API 的路径可能不带 /v1
+            if resp.status_code == 404:
+                url_alt = (base_url.rstrip('/') if base_url else 'https://api.openai.com') + '/models'
+                resp = http_requests.get(url_alt, headers=headers, timeout=15)
+
+            if resp.status_code != 200:
+                return jsonify({'error': f'API 返回 {resp.status_code}: {resp.text[:200]}'}), 502
+
+            body = resp.json()
+            models = sorted([m['id'] for m in body.get('data', [])]) if 'data' in body else []
+            return jsonify({'models': models})
+
+        elif provider_type == 'anthropic':
+            # Anthropic 没有官方 list models 接口，返回常用模型列表
+            models = [
+                'claude-opus-4-20250514',
+                'claude-sonnet-4-20250514',
+                'claude-haiku-4-20250506',
+                'claude-3-5-sonnet-20241022',
+                'claude-3-5-haiku-20241022',
+                'claude-3-opus-20240229',
+            ]
+            return jsonify({'models': models})
+
+        else:
+            return jsonify({'error': f'不支持的 provider 类型: {provider_type}'}), 400
+
+    except http_requests.Timeout:
+        return jsonify({'error': '请求超时，请检查 Base URL 是否正确'}), 504
+    except http_requests.ConnectionError:
+        return jsonify({'error': '无法连接目标 API，请检查 Base URL'}), 502
+    except Exception as e:
+        logger.exception("获取模型列表失败")
+        return jsonify({'error': str(e)}), 500
+
+
+@bp.route('/paddleocr/test', methods=['POST'])
+def test_paddleocr():
+    """测试 PaddleOCR API Token 是否可用。
+
+    请求体:
+        api_token: API Token（可选，留空则从数据库读取）
+        api_url: API URL（可选，留空则从数据库读取）
+        provider_id: 已有 provider 的 id（可选，用于回退读取已存 token）
+    """
+    try:
+        data = request.get_json(force=True) or {}
+        api_token = data.get('api_token') or ''
+        api_url = data.get('api_url') or ''
+
+        # 前端未提供凭据时，从数据库读取已保存的配置
+        if not api_token or not api_url:
+            user_id = session.get('user_id')
+            provider_id = data.get('provider_id')
+            if user_id:
+                with SessionLocal() as db:
+                    if provider_id:
+                        provider = db.query(ProviderConfig).filter_by(
+                            id=provider_id, user_id=user_id
+                        ).first()
+                    else:
+                        provider = crud.get_active_provider(db, user_id, 'paddleocr')
+                    if provider:
+                        if not api_token:
+                            api_token = provider.api_key or ''
+                        if not api_url:
+                            api_url = provider.base_url or ''
+
+        if not api_token or not api_url:
+            return jsonify({'error': '未提供 API Token 或 API URL'}), 400
+
+        # 用 GET 请求 API URL，带 token 验证认证是否有效
+        headers = {'Authorization': f'bearer {api_token}'}
+        resp = http_requests.get(api_url, headers=headers, timeout=10)
+
+        # 401/403 表示 token 无效
+        if resp.status_code in (401, 403):
+            return jsonify({'success': False, 'error': '认证失败，请检查 API Token 是否正确'})
+
+        # 其他非 5xx 状态码都认为 token 有效（包括 404/405 等，因为 GET 可能不被支持但认证通过了）
+        if resp.status_code >= 500:
+            return jsonify({'success': False, 'error': f'服务端错误 ({resp.status_code})，请检查 API URL'})
+
+        return jsonify({'success': True, 'message': 'API Token 验证通过'})
+
+    except http_requests.Timeout:
+        return jsonify({'success': False, 'error': '请求超时，请检查 API URL 是否正确'}), 504
+    except http_requests.ConnectionError:
+        return jsonify({'success': False, 'error': '无法连接，请检查 API URL'}), 502
+    except Exception as e:
+        logger.exception("测试 PaddleOCR 连接失败")
+        return jsonify({'success': False, 'error': str(e)}), 500
diff --git a/backend/routes/stats.py b/backend/routes/stats.py
new file mode 100644
index 00000000..b74f5f7c
--- /dev/null
+++ b/backend/routes/stats.py
@@ -0,0 +1,79 @@
+import logging
+
+from flask import Blueprint, request, jsonify, session
+
+from db import SessionLocal
+from db import crud
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('stats', __name__)
+
+
+@bp.route('/stats', methods=['GET'])
+def get_stats():
+    """
+    获取知识点统计信息
+    """
+    try:
+        subject = request.args.get('subject', type=str) or None
+        with SessionLocal() as db:
+            stats = crud.get_knowledge_stats(db, subject=subject, user_id=session.get('user_id'))
+            return jsonify({
+                'success': True,
+                'stats': stats,
+                'total_tags': len(stats)
+            })
+
+    except Exception as e:
+        logger.exception("获取统计失败")
+        return jsonify({'success': False, 'error': '获取统计失败，请稍后重试'}), 500
+
+
+@bp.route('/dashboard-stats', methods=['GET'])
+def get_dashboard_stats():
+    """获取 Dashboard 所需的完整统计数据，支持 ?subject= 学科筛选"""
+    try:
+        subject = request.args.get('subject') or None
+        uid = session.get('user_id')
+        with SessionLocal() as db:
+            # 可用学科列表（按当前用户隔离）
+            subjects = crud.get_existing_subjects(db, user_id=uid)
+
+            # 复习状态统计
+            review_stats = crud.get_review_status_stats(db, subject=subject, user_id=uid)
+
+            # 总体统计
+            statistics = crud.get_statistics(db, subject=subject, user_id=uid)
+
+            # 知识点标签统计 top 10（横向条形图）
+            tag_stats = crud.get_knowledge_stats(db, subject=subject, limit=10, user_id=uid)
+
+            # 知识点 × 掌握状态（堆叠柱状图）
+            tag_status_stats = crud.get_tag_status_stats(db, subject=subject, limit=10, user_id=uid)
+
+            # 知识点 × 题型（热力图）
+            tag_type_stats = crud.get_tag_type_stats(db, subject=subject, tag_limit=8, user_id=uid)
+
+            # 今日掌握数
+            today_mastered = crud.get_today_mastered_count(db, subject=subject, user_id=uid)
+
+            # 最近30天每日新增 + 已掌握趋势
+            daily_counts = crud.get_daily_counts(db, days=30, subject=subject, user_id=uid)
+
+            return jsonify({
+                'success': True,
+                'subjects': subjects,
+                'review_stats': review_stats,
+                'today_mastered': today_mastered,
+                'total_questions': statistics['total_questions'],
+                'by_subject': statistics['by_subject'],
+                'tag_stats': tag_stats,
+                'tag_status_stats': tag_status_stats,
+                'tag_type_stats': tag_type_stats,
+                'daily_counts': daily_counts,
+            })
+
+    except Exception as e:
+        logger.exception("获取 Dashboard 统计失败")
+        return jsonify({'success': False, 'error': '获取统计失败，请稍后重试'}), 500
diff --git a/backend/routes/upload.py b/backend/routes/upload.py
new file mode 100644
index 00000000..357a94d5
--- /dev/null
+++ b/backend/routes/upload.py
@@ -0,0 +1,648 @@
+import os
+import json
+import uuid
+import logging
+from pathlib import PurePath
+from typing import Optional
+
+from flask import Blueprint, request, jsonify, session
+
+import core.state as state
+from core.state import (
+    workflow_graph,
+    session_lock,
+    get_user_session,
+)
+from db import SessionLocal
+from db import crud
+from core.config import settings
+from src.utils import export_wrongbook as export_wrongbook_md
+
+logger = logging.getLogger(__name__)
+
+bp = Blueprint('upload', __name__)
+
+
+# ---------------------------------------------------------------------------
+# Helper functions
+# ---------------------------------------------------------------------------
+
+def allowed_file(filename):
+    """检查文件扩展名是否允许"""
+    return PurePath(filename).suffix.lower().lstrip('.') in settings.allowed_extensions
+
+
+def _read_split_subject() -> Optional[str]:
+    """从 split_metadata.json 读取学科信息"""
+    meta_path = os.path.join(str(settings.results_dir), "split_metadata.json")
+    if os.path.exists(meta_path):
+        with open(meta_path, 'r', encoding='utf-8') as f:
+            return json.load(f).get("subject")
+    return None
+
+
+def _serialize_split_record(r) -> dict:
+    """将 SplitRecord ORM 对象序列化为前端 JSON 格式"""
+    return {
+        "id": r.id,
+        "subject": r.subject,
+        "model_provider": r.model_provider,
+        "file_names": json.loads(r.file_names_json) if r.file_names_json else [],
+        "question_count": r.question_count,
+        "created_at": r.created_at.isoformat() if r.created_at else None,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Routes
+# ---------------------------------------------------------------------------
+
+@bp.route('/upload', methods=['POST'])
+def upload_file():
+    """处理文件上传（支持多文件）。
+
+    这里只负责把原始文件保存到 uploads 并登记到会话中；
+    标准化(PDF/图片 → 图片列表) + OCR + 分割，会在 /api/split 里由用户点击"开始分割"后统一触发。
+
+    Returns:
+        JSON响应，包含上传结果
+    """
+    # 支持多文件：前端用 'files' 字段发送，兼容单文件 'file' 字段
+    files = request.files.getlist('files')
+    if not files:
+        files = request.files.getlist('file')
+    if not files or all(f.filename == '' for f in files):
+        return jsonify({'error': '没有上传文件'}), 400
+
+    # 校验每个文件
+    for file in files:
+        if file.filename == '':
+            continue
+
+        if not allowed_file(file.filename):
+            return jsonify({
+                'error': f'不支持的文件格式: {file.filename}。支持: {", ".join(settings.allowed_extensions)}'
+            }), 400
+
+        file.seek(0, 2)
+        file_size = file.tell()
+        file.seek(0)
+        file_size_mb = file_size / (1024 * 1024)
+
+        if file_size_mb > settings.max_file_size_mb:
+            return jsonify({
+                'error': f'{file.filename} 大小为 {file_size_mb:.1f}MB，超出最大限制 {settings.max_file_size_mb}MB'
+            }), 400
+
+        if file_size == 0:
+            return jsonify({
+                'error': f'{file.filename} 为空文件，请重新选择'
+            }), 400
+
+    try:
+        file_keys = request.form.getlist('file_key')
+        if not file_keys:
+            file_keys = request.form.getlist('file_keys')
+
+        prepared = []
+        for i, file in enumerate(files):
+            if file.filename == '':
+                continue
+            fk = file_keys[i] if i < len(file_keys) and file_keys[i] else None
+            prepared.append((fk, file))
+
+        if not prepared:
+            return jsonify({'error': '没有上传文件'}), 400
+
+        results_out = []
+        uid = session.get('user_id')
+        for fk, file in prepared:
+            file_key = fk or f"{uuid.uuid4().hex}"
+
+            original_ext = os.path.splitext(file.filename)[1].lower().lstrip('.')
+            filename = f"{uuid.uuid4().hex}.{original_ext}"
+            filepath = os.path.join(str(settings.upload_dir), filename)
+            file.save(filepath)
+
+            with session_lock:
+                us = get_user_session(uid)
+                cancelled = file_key in us["cancelled_file_keys"]
+                if cancelled:
+                    us["cancelled_file_keys"].discard(file_key)
+
+            if cancelled:
+                try:
+                    os.remove(filepath)
+                except FileNotFoundError:
+                    pass
+                continue
+
+            with session_lock:
+                us = get_user_session(uid)
+                if us["current_thread_id"] is not None:
+                    us["current_thread_id"] = None
+                    us["session_files"].clear()
+                    us["session_file_order"].clear()
+
+                us["session_files"][file_key] = {
+                    "filename": file.filename,
+                    "filepath": filepath,
+                }
+                if file_key not in us["session_file_order"]:
+                    us["session_file_order"].append(file_key)
+
+            results_out.append({
+                "file_key": file_key,
+                "filename": file.filename,
+            })
+
+        return jsonify({
+            'success': True,
+            'message': '上传成功',
+            'result': {
+                'file_count': len(results_out),
+                'files': results_out,
+            }
+        })
+
+    except FileNotFoundError:
+        return jsonify({
+            'success': False,
+            'error': '文件保存失败，请重新上传'
+        }), 500
+    except Exception as e:
+        logger.exception("文件处理失败")
+        return jsonify({
+            'success': False,
+            'error': '文件处理失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/cancel_file', methods=['POST'])
+def cancel_file():
+    try:
+        data = request.get_json(silent=True) or {}
+        file_key = data.get('file_key')
+        if not file_key:
+            return jsonify({
+                'success': False,
+                'error': '缺少 file_key'
+            }), 400
+
+        uid = session.get('user_id')
+        with session_lock:
+            us = get_user_session(uid)
+            if us["current_thread_id"] is not None:
+                return jsonify({
+                    'success': False,
+                    'error': '已开始分割，无法撤销单个文件；请重置后重新上传'
+                }), 400
+
+        filepath = None
+        existed = False
+        with session_lock:
+            us = get_user_session(uid)
+            us["cancelled_file_keys"].add(file_key)
+
+            existed = file_key in us["session_files"]
+            v = us["session_files"].pop(file_key, None) or {}
+            filepath = v.get('filepath')
+            if file_key in us["session_file_order"]:
+                us["session_file_order"].remove(file_key)
+
+            us["cancelled_file_keys"].discard(file_key)
+
+        if filepath:
+            try:
+                os.remove(filepath)
+            except FileNotFoundError:
+                pass
+
+        return jsonify({
+            'success': True,
+            'message': '已撤销该文件' if existed else '已标记撤销该文件',
+        })
+
+    except Exception as e:
+        logger.exception("撤销文件失败")
+        return jsonify({
+            'success': False,
+            'error': '撤销失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/erase', methods=['POST'])
+def run_erase():
+    """擦除手写笔迹，返回擦除后的图片 URL。
+
+    擦除后的文件路径保存在会话状态中，供后续 /api/ocr 使用。
+    """
+    try:
+        uid = session.get('user_id')
+        with session_lock:
+            us = get_user_session(uid)
+            keys = list(us["session_file_order"])
+            file_paths = []
+            for k in keys:
+                v = us["session_files"].get(k) or {}
+                fp = v.get('filepath')
+                if fp:
+                    file_paths.append(fp)
+
+        if not file_paths:
+            return jsonify({'success': False, 'error': '请先上传文件'}), 400
+
+        ensexam_configured = settings.model_path.exists()
+        if not ensexam_configured:
+            return jsonify({'success': False, 'error': 'EnsExam 模型未配置'}), 400
+
+        from models.inference import InferenceEngine
+        engine = InferenceEngine()
+        erased_paths = []
+        for fp in file_paths:
+            with open(fp, 'rb') as f:
+                img_bytes = f.read()
+            result_img = engine.run(img_bytes)
+            erased_name = f"auto_{uuid.uuid4().hex[:8]}_{os.path.splitext(os.path.basename(fp))[0]}.png"
+            erased_path = settings.erased_dir / erased_name
+            result_img.save(str(erased_path), format='PNG')
+            erased_paths.append(str(erased_path))
+
+        # 保存擦除后的路径到会话，供 /api/ocr 复用
+        with session_lock:
+            us = get_user_session(uid)
+            us["erased_file_paths"] = erased_paths
+
+        preview = []
+        for i, (orig, erased) in enumerate(zip(file_paths, erased_paths)):
+            preview.append({
+                "index": i,
+                "original_url": f"/api/image/{os.path.basename(orig)}",
+                "erased_url": f"/api/image/{os.path.basename(erased)}",
+            })
+
+        return jsonify({
+            'success': True,
+            'message': f'擦除完成，共 {len(erased_paths)} 张图片',
+            'images': preview,
+        })
+
+    except Exception:
+        logger.exception("擦除手写笔迹失败")
+        return jsonify({'success': False, 'error': '擦除失败，请稍后重试'}), 500
+
+
+@bp.route('/ocr', methods=['POST'])
+def run_ocr():
+    """只执行 OCR，返回识别结果供用户预览。
+
+    流程：标准化输入 → OCR 解析 → 返回结构化文本
+    不触发 Agent 分割。
+    如果之前调用过 /api/erase，则使用擦除后的图片。
+    """
+    try:
+        # 加载 OCR 凭据
+        user_id = session.get('user_id')
+        ocr_credentials = {}
+        if user_id:
+            with SessionLocal() as db:
+                ocr_provider = crud.get_active_provider(db, user_id, 'paddleocr')
+                if ocr_provider:
+                    ocr_credentials = {
+                        "api_url": ocr_provider.base_url,
+                        "token": ocr_provider.api_key,
+                        "model": ocr_provider.model_name,
+                        "use_doc_orientation": ocr_provider.use_doc_orientation,
+                        "use_doc_unwarping": ocr_provider.use_doc_unwarping,
+                        "use_chart_recognition": ocr_provider.use_chart_recognition,
+                    }
+
+        # 优先使用擦除后的路径
+        file_paths = None
+        with session_lock:
+            us = get_user_session(user_id)
+            if us["erased_file_paths"]:
+                file_paths = list(us["erased_file_paths"])
+                us["erased_file_paths"] = None  # 用完即清
+
+        if not file_paths:
+            with session_lock:
+                us = get_user_session(user_id)
+                keys = list(us["session_file_order"])
+                file_paths = []
+                for k in keys:
+                    v = us["session_files"].get(k) or {}
+                    fp = v.get('filepath')
+                    if fp:
+                        file_paths.append(fp)
+
+        if not file_paths:
+            return jsonify({'success': False, 'error': '请先上传文件'}), 400
+
+        # 标准化输入（PDF → 图片）
+        from src.utils import prepare_input
+        all_image_paths = []
+        for fp in file_paths:
+            all_image_paths.extend(prepare_input(fp))
+
+        # 执行 OCR（获取原始结果用于预览 bbox，简化结果用于后续分割）
+        from src.paddleocr_client import PaddleOCRClient
+        from src.utils import simplify_ocr_results
+        from src.workflow import run_async
+
+        creds = ocr_credentials or {}
+        client = PaddleOCRClient(
+            api_url=creds.get("api_url"),
+            token=creds.get("token"),
+            model=creds.get("model"),
+            use_doc_orientation=creds.get("use_doc_orientation"),
+            use_doc_unwarping=creds.get("use_doc_unwarping"),
+            use_chart_recognition=creds.get("use_chart_recognition"),
+        )
+
+        # 按文件类型分组
+        pdf_paths = [p for p in all_image_paths if p.lower().endswith(".pdf")]
+        image_only = [p for p in all_image_paths if not p.lower().endswith(".pdf")]
+
+        raw_ocr_results = []
+        for pdf_path in pdf_paths:
+            try:
+                raw_ocr_results.append(client.parse_pdf(pdf_path, save_output=True))
+            except Exception:
+                logger.exception(f"OCR PDF 失败: {pdf_path}")
+        if image_only:
+            try:
+                img_results = run_async(client.parse_images_async(image_only, save_output=True))
+                if img_results:
+                    raw_ocr_results.extend(img_results)
+            except Exception:
+                logger.exception("OCR 图片失败")
+
+        if not raw_ocr_results:
+            return jsonify({'success': False, 'error': 'OCR 解析失败，请检查 PaddleOCR 配置'}), 500
+
+        # 简化结果用于后续分割
+        ocr_data = simplify_ocr_results(raw_ocr_results)
+
+        # 保存 OCR 结果到会话，供后续 /api/split 复用
+        import json
+        results_dir = str(settings.results_dir)
+        os.makedirs(results_dir, exist_ok=True)
+        ocr_cache_path = os.path.join(results_dir, "ocr_cache.json")
+        with open(ocr_cache_path, 'w', encoding='utf-8') as f:
+            json.dump(ocr_data, f, ensure_ascii=False, indent=2)
+
+        # 构建前端预览数据：每页的图片 + bbox 标注
+        preview_pages = []
+        page_idx = 0
+        for result in raw_ocr_results:
+            for layout_page in result.get("layoutParsingResults", []):
+                pruned = layout_page.get("prunedResult", {})
+                page_w = pruned.get("width", 1)
+                page_h = pruned.get("height", 1)
+                parsing_list = pruned.get("parsing_res_list", [])
+
+                blocks = []
+                for b in parsing_list:
+                    bbox = b.get("block_bbox")
+                    if not bbox:
+                        continue
+                    blocks.append({
+                        "label": b.get("block_label", ""),
+                        "content": (b.get("block_content", "") or "")[:80],
+                        "bbox": bbox,
+                    })
+
+                image_path = all_image_paths[page_idx] if page_idx < len(all_image_paths) else None
+                image_url = f"/api/image/{os.path.basename(image_path)}" if image_path else None
+
+                preview_pages.append({
+                    "page_index": page_idx,
+                    "image_url": image_url,
+                    "page_width": page_w,
+                    "page_height": page_h,
+                    "blocks": blocks,
+                })
+                page_idx += 1
+
+        return jsonify({
+            'success': True,
+            'message': f'OCR 完成，共 {len(preview_pages)} 页',
+            'pages': preview_pages,
+            'total_blocks': sum(len(p["blocks"]) for p in preview_pages),
+        })
+
+    except Exception:
+        logger.exception("OCR 执行失败")
+        return jsonify({'success': False, 'error': 'OCR 执行失败，请稍后重试'}), 500
+
+
+@bp.route('/split', methods=['POST'])
+def split_questions():
+    """开始执行标准化 + OCR + 分割。
+
+    用户点击前端"开始分割题目"后调用该接口：
+    - 标准化输入（PDF/图片 → 图片列表）
+    - 触发 Agent/OCR 并分割题目
+
+    Returns:
+        JSON响应，包含分割后的题目
+    """
+    try:
+        # 读取请求体参数（模型供应商 + 模型名称）
+        data = request.get_json(silent=True) or {}
+        model_provider = data.get("model_provider", "openai")
+        model_name = data.get("model_name")  # 可选，None 时使用 provider 默认模型
+        if not settings.is_valid_provider(model_provider):
+            return jsonify({
+                'success': False,
+                'error': f'不支持的模型供应商: {model_provider}'
+            }), 400
+
+        # 从数据库加载用户的 LLM + OCR 凭据
+        user_id = session.get('user_id')
+        ocr_credentials = {}
+        if user_id:
+            settings.load_providers_from_db(user_id)
+            with SessionLocal() as db:
+                ocr_provider = crud.get_active_provider(db, user_id, 'paddleocr')
+                if ocr_provider:
+                    ocr_credentials = {
+                        "api_url": ocr_provider.base_url,
+                        "token": ocr_provider.api_key,
+                        "model": ocr_provider.model_name,
+                        "use_doc_orientation": ocr_provider.use_doc_orientation,
+                        "use_doc_unwarping": ocr_provider.use_doc_unwarping,
+                        "use_chart_recognition": ocr_provider.use_chart_recognition,
+                    }
+
+        with session_lock:
+            us = get_user_session(user_id)
+            keys = list(us["session_file_order"])
+            file_paths = []
+            for k in keys:
+                v = us["session_files"].get(k) or {}
+                fp = v.get('filepath')
+                if fp:
+                    file_paths.append(fp)
+
+        if not file_paths:
+            return jsonify({
+                'success': False,
+                'error': '请先上传文件'
+            }), 400
+
+        with session_lock:
+            us = get_user_session(user_id)
+            us["current_thread_id"] = str(uuid.uuid4())
+            thread_id = us["current_thread_id"]
+        config = {"configurable": {"thread_id": thread_id}}
+
+        initial_state = {
+            "file_paths": file_paths,
+            "model_provider": model_provider,
+            "model_name": model_name,
+            "ocr_credentials": ocr_credentials,
+        }
+        workflow_graph.invoke(initial_state, config=config)
+        result_state = workflow_graph.invoke(None, config=config)
+
+        questions = result_state.get('questions', [])
+        warnings = result_state.get('warnings', [])
+
+        # 自动保存分割记录
+        try:
+            subject = _read_split_subject()
+
+            with session_lock:
+                us = get_user_session(user_id)
+                file_names = [
+                    us["session_files"].get(k, {}).get("filename", "未知")
+                    for k in us["session_file_order"]
+                ]
+
+            with SessionLocal() as db:
+                crud.save_split_record(db, subject, model_provider, file_names, questions, user_id=user_id)
+        except Exception:
+            logger.warning("保存分割记录失败，不影响主流程", exc_info=True)
+
+        return jsonify({
+            'success': True,
+            'message': f'成功分割 {len(questions)} 道题目',
+            'questions': questions,
+            'warnings': warnings,
+        })
+
+    except Exception as e:
+        logger.exception("题目分割失败")
+        return jsonify({
+            'success': False,
+            'error': '题目分割失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/split-records', methods=['GET'])
+def get_split_records():
+    """获取最近 N 次分割历史记录，limit 由前端通过查询参数指定"""
+    try:
+        limit = request.args.get('limit', 10, type=int)
+        limit = max(1, min(limit, crud.MAX_SPLIT_RECORDS))  # 上限与保留条数一致
+
+        with SessionLocal() as db:
+            records = crud.get_recent_split_records(db, limit, user_id=session.get('user_id'))
+            result = [_serialize_split_record(r) for r in records]
+
+        return jsonify({"success": True, "records": result})
+
+    except Exception as e:
+        logger.exception("获取分割记录失败")
+        return jsonify({"success": False, "error": "获取分割记录失败"}), 500
+
+
+@bp.route('/split-records/<int:record_id>', methods=['GET'])
+def get_split_record_detail(record_id):
+    """获取单条分割记录的完整数据（含 questions）"""
+    try:
+        with SessionLocal() as db:
+            uid = None if session.get('is_admin') else session.get('user_id')
+            record = crud.get_split_record_by_id(db, record_id, user_id=uid)
+            if not record:
+                return jsonify({"success": False, "error": "记录不存在"}), 404
+
+            result = _serialize_split_record(record)
+            result["questions"] = json.loads(record.questions_json) if record.questions_json else []
+
+        return jsonify({"success": True, "record": result})
+
+    except Exception as e:
+        logger.exception("获取分割记录详情失败")
+        return jsonify({"success": False, "error": "获取分割记录详情失败"}), 500
+
+
+@bp.route('/export', methods=['POST'])
+def export_wrongbook():
+    """
+    注入选中题目 ID 并恢复图执行导出
+
+    图执行: export → END
+
+    Returns:
+        JSON响应，包含导出文件路径
+    """
+    try:
+        data = request.get_json(silent=True) or {}
+        selected_uids = data.get('selected_ids', [])
+
+        if not isinstance(selected_uids, list):
+            return jsonify({
+                'success': False,
+                'error': 'selected_ids 必须为列表'
+            }), 400
+
+        if not selected_uids:
+            return jsonify({
+                'success': False,
+                'error': '未选择任何题目'
+            }), 400
+
+        # 检查是否已分割（通过 questions.json 存在性判断，不依赖内存中的 current_thread_id）
+        results_dir = settings.results_dir
+        questions_file = os.path.join(results_dir, "questions.json")
+        if not os.path.exists(questions_file):
+            return jsonify({
+                'success': False,
+                'error': '请先分割题目'
+            }), 400
+
+        with open(questions_file, 'r', encoding='utf-8') as f:
+            questions = json.load(f)
+
+        output_path = export_wrongbook_md(
+            questions,
+            selected_uids,
+        )
+
+        return jsonify({
+            'success': True,
+            'message': '错题本导出成功',
+            'output_path': output_path
+        })
+
+    except Exception as e:
+        logger.exception("导出失败")
+        return jsonify({
+            'success': False,
+            'error': '导出失败，请稍后重试'
+        }), 500
+
+
+@bp.route('/image/<filename>')
+def serve_image(filename):
+    """提供上传/处理后的图片文件访问（OCR 预览用）"""
+    from flask import send_from_directory
+    # 在多个目录中查找
+    for d in [settings.upload_dir, settings.erased_dir, settings.results_dir, settings.pages_dir]:
+        path = os.path.join(str(d), filename)
+        if os.path.exists(path):
+            return send_from_directory(str(d), filename)
+    return jsonify({'error': 'File not found'}), 404
diff --git a/backend/src/paddleocr_client.py b/backend/src/paddleocr_client.py
index af16130f..79b13db3 100644
--- a/backend/src/paddleocr_client.py
+++ b/backend/src/paddleocr_client.py
@@ -120,7 +120,7 @@ def _parse_file(
     ) -> Dict[str, Any]:
         """解析文件并返回结构化结果（同步，任务轮询模式）"""
         if output_dir is None:
-            from config import settings
+            from core.config import settings
             output_dir = settings.struct_dir
 
         console.print(f"[cyan]正在解析: {file_path}[/cyan]")
@@ -270,7 +270,7 @@ async def async_parse_image(
     ) -> Dict[str, Any]:
         """异步解析单张图片（提交任务 → 轮询 → 下载结果）"""
         if output_dir is None:
-            from config import settings
+            from core.config import settings
             output_dir = settings.struct_dir
 
         console.print(f"[cyan]正在解析图片: {image_path}[/cyan]")
diff --git a/backend/src/utils.py b/backend/src/utils.py
index 6d5f07bd..db8ad164 100644
--- a/backend/src/utils.py
+++ b/backend/src/utils.py
@@ -4,6 +4,7 @@
 """
 
 import os
+import re
 import shutil
 import asyncio
 from concurrent.futures import ThreadPoolExecutor
@@ -12,9 +13,9 @@
 from dotenv import load_dotenv
 from rich.console import Console
 
-from config import settings
+from core.config import settings
 
-load_dotenv()
+load_dotenv(os.path.join(os.path.dirname(__file__), "..", ".env"))
 console = Console()
 
 
@@ -62,7 +63,7 @@ def prepare_input(file_path: str) -> List[str]:
 
 def export_wrongbook(
     questions: List[Dict[str, Any]],
-    selected_ids: List[str],
+    selected_uids: List[str],
     output_path: str = None,
     batch_info: Dict[str, Any] = None
 ) -> str:
@@ -73,7 +74,7 @@ def export_wrongbook(
 
     Args:
         questions: 题目列表
-        selected_ids: 用户选择的题目ID列表
+        selected_uids: 用户选择的题目 uid 列表
         output_path: 输出Markdown文件路径（可选）
         batch_info: 批次信息（用于入库），包含 original_filename, subject 等
 
@@ -84,59 +85,107 @@ def export_wrongbook(
         os.makedirs(settings.results_dir, exist_ok=True)
         output_path = os.path.join(settings.results_dir, "wrongbook.md")
 
-    # 过滤选中的题目
-    selected_questions = [q for q in questions if q.get('question_id') in selected_ids]
+    # 用 uid 过滤选中的题目，保持原始顺序
+    uid_set = set(selected_uids)
+    selected_questions = [q for q in questions if q.get('uid') in uid_set]
 
     # 构建Markdown内容
     md_content = "# 错题本\n\n"
     md_content += f"> 共收录 {len(selected_questions)} 道题目\n\n"
     md_content += "---\n\n"
 
-    for i, q in enumerate(selected_questions, start=1):
-        md_content += f"## {i}. 题目 {q.get('question_id', '')} ({q.get('question_type', '未知')})\n\n"
+    rel_struct_dir = os.path.relpath(settings.struct_dir, settings.results_dir)
+    _img_src_re = re.compile(r'src="((?:/images/|imgs/)[^"]+)"')
+
+    def _resolve_image_path(p: str) -> str:
+        """将 Agent 输出的图片路径统一转为 Markdown 可用的相对路径。
+
+        Agent 有时输出 '/images/xxx.jpg'（Flask 路由），有时输出 'imgs/xxx.jpg'
+        （LLM 对路径的自由变形），统一映射到 struct_dir/imgs/ 下的相对路径。
+        """
+        p = p.strip()
+        if p.startswith("/images/"):
+            return f"{rel_struct_dir}/imgs/{p[len('/images/'):]}"
+        if p.startswith("imgs/"):
+            return f"{rel_struct_dir}/imgs/{p[len('imgs/'):]}"
+        return p
+
+    def _fix_html_image_src(html: str) -> str:
+        """修正 HTML 内 <img src> 中的图片路径，与独立 image block 保持一致。"""
+        return _img_src_re.sub(
+            lambda m: f'src="{_resolve_image_path(m.group(1))}"', html
+        )
+
+    _INIT = object()  # 初始哨兵，确保第一个有 section_title 的题一定触发标题输出
+    current_section = _INIT
+    unsorted_started = False
+    serial = 0  # 全局序号
+
+    for q in selected_questions:
+        section = q.get('section_title')
+
+        if section:
+            # 有 section_title：正常输出大题分组标题
+            if section != current_section:
+                current_section = section
+                md_content += f"## {section}\n\n"
+        else:
+            # section=None：首次遇到时输出"（未分类）"节标题
+            if not unsorted_started:
+                unsorted_started = True
+                md_content += "## （未分类）\n\n"
+
+        serial += 1
+        md_content += f"### {serial}. 题目 {q.get('question_id', '')} ({q.get('question_type', '未知')})\n\n"
 
         # 获取图片引用列表，用于填充空的 image block
         image_refs = q.get('image_refs') or []
-        rendered_images = set()  # 记录已渲染的图片路径
+        rendered_images = set()  # 记录已渲染的图片路径（原始路径，供 image_refs 兜底比对）
+
+        # 预扫描：收集已内嵌在 HTML 文本块（如 table）中的图片路径，避免重复渲染
+        html_embedded = set()
+        for block in q.get('content_blocks', []):
+            if block.get('block_type') == 'text':
+                for m in _img_src_re.finditer(block.get('content', '')):
+                    html_embedded.add(m.group(1).strip())
 
         # 添加内容块（只有 text 和 image，公式以 LaTeX 标记嵌入 text 中）
         for block in q.get('content_blocks', []):
             if block['block_type'] == 'text':
-                md_content += f"{block['content']}\n\n"
+                content = block['content'].replace('\n', '  \n')
+                # 修正 HTML 内嵌图片路径（table 中的 <img src>）
+                content = _fix_html_image_src(content)
+                md_content += f"{content}\n\n"
             elif block['block_type'] == 'image':
                 image_path = block.get('content', '').strip()
-                if image_path:
-                    rendered_images.add(image_path)
-
-                # 将 Flask 路由路径转为 Markdown 相对路径
-                if image_path.startswith("/images/"):
-                    rel_struct_dir = os.path.relpath(settings.struct_dir, settings.results_dir)
-                    image_path = f"{rel_struct_dir}/imgs/{image_path[len('/images/') :]}"
-
-                if image_path:
-                    md_content += f"![图片]({image_path})\n\n"
+                if not image_path or image_path in html_embedded:
+                    # 已在 HTML 表格中渲染过，跳过独立 image block 避免重复
+                    continue
+                rendered_images.add(image_path)
+                resolved = _resolve_image_path(image_path)
+                if resolved:
+                    md_content += f"![图片]({resolved})\n\n"
 
         # 添加选项
         if q.get('options'):
             for option in q['options']:
-                md_content += f"{option}\n\n"
+                md_content += f"{option}  \n"
+            md_content += "\n"
 
-        # 兜底：渲染 image_refs 中未被 content_blocks 覆盖的图片
-        remaining_images = [p for p in image_refs if p not in rendered_images]
+        # 兜底：渲染 image_refs 中未被 content_blocks 覆盖、也未在 HTML 中出现的图片
+        remaining_images = [p for p in image_refs if p not in rendered_images and p not in html_embedded]
         if remaining_images:
             for image_path in remaining_images:
-                if image_path.startswith("/images/"):
-                    rel_struct_dir = os.path.relpath(settings.struct_dir, settings.results_dir)
-                    image_path = f"{rel_struct_dir}/imgs/{image_path[len('/images/') :]}"
-                md_content += f"![图片]({image_path})\n\n"
+                md_content += f"![图片]({_resolve_image_path(image_path)})\n\n"
 
-        md_content += "### 我的答案\n\n"
+        answer_prefix = "####" if section else "###"
+        md_content += f"{answer_prefix} 我的答案\n\n"
         md_content += "_（请在此处填写你的答案）_\n\n"
 
-        md_content += "### 正确答案\n\n"
+        md_content += f"{answer_prefix} 正确答案\n\n"
         md_content += "_（请在此处填写正确答案）_\n\n"
 
-        md_content += "### 解析\n\n"
+        md_content += f"{answer_prefix} 解析\n\n"
         md_content += "_（请在此处填写解题思路和知识点）_\n\n"
 
         md_content += "---\n\n"
@@ -185,6 +234,9 @@ def simplify_ocr_results(ocr_results: list) -> List[Dict[str, Any]]:
                     if bbox:
                         prefix = "img_in_chart_box" if label == "chart" else "img_in_image_box"
                         content = f"/images/{prefix}_{int(bbox[0])}_{int(bbox[1])}_{int(bbox[2])}_{int(bbox[3])}.jpg"
+                # 统一 OCR 原始输出中的图片路径：imgs/ → /images/
+                if 'src="imgs/' in content:
+                    content = content.replace('src="imgs/', 'src="/images/')
                 slim_blocks.append({
                     "block_label": label,
                     "block_content": content,
diff --git a/backend/src/workflow.py b/backend/src/workflow.py
index ce0c8f7d..4ed8c1e7 100644
--- a/backend/src/workflow.py
+++ b/backend/src/workflow.py
@@ -4,16 +4,21 @@
 """
 
 import os
+import re as _re
 import json
 import logging
 import time
+import traceback
+import difflib as _difflib
 from typing import List, Dict, Any, TypedDict
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from rich.console import Console
 from langgraph.graph import StateGraph, START, END
 from langgraph.checkpoint.memory import MemorySaver
-from config import settings
+from core.config import settings
 from .utils import prepare_input, export_wrongbook, simplify_ocr_results, run_async
+
+_HTML_TAG_RE = _re.compile(r"<[^>]+>")
 console = Console()
 
 # 配置日志
@@ -106,14 +111,15 @@ def _run_ocr_and_simplify(file_paths: List[str], ocr_credentials: Dict[str, Any]
                 break
             except Exception as e:
                 last_error = e
+                tb = traceback.format_exc()
                 if attempt < max_retries:
                     delay = retry_delays[attempt - 1]
-                    logger.warning(f"OCR PDF 第 {attempt} 次失败 ({e})，{delay}s 后重试...")
-                    console.print(f"[yellow]OCR PDF 第 {attempt} 次失败，{delay}s 后重试...[/yellow]")
+                    logger.warning(f"OCR PDF 第 {attempt} 次失败 ({type(e).__name__}: {e})，{delay}s 后重试...\n{tb}")
+                    console.print(f"[yellow]OCR PDF 第 {attempt} 次失败 ({type(e).__name__}: {e})，{delay}s 后重试...[/yellow]")
                     time.sleep(delay)
                 else:
-                    logger.error(f"OCR PDF 全部 {max_retries} 次重试失败: {last_error}")
-                    console.print(f"[red]OCR PDF 解析失败（已重试 {max_retries} 次）: {last_error}[/red]")
+                    logger.error(f"OCR PDF 全部 {max_retries} 次重试失败 ({type(last_error).__name__}: {last_error})\n{tb}")
+                    console.print(f"[red]OCR PDF 解析失败（已重试 {max_retries} 次）: {type(last_error).__name__}: {last_error}[/red]")
         if result is not None:
             ocr_results.append(result)
 
@@ -128,14 +134,15 @@ def _run_ocr_and_simplify(file_paths: List[str], ocr_credentials: Dict[str, Any]
                 break
             except Exception as e:
                 last_error = e
+                tb = traceback.format_exc()
                 if attempt < max_retries:
                     delay = retry_delays[attempt - 1]
-                    logger.warning(f"OCR 图片第 {attempt} 次失败 ({e})，{delay}s 后重试...")
-                    console.print(f"[yellow]OCR 图片第 {attempt} 次失败，{delay}s 后重试...[/yellow]")
+                    logger.warning(f"OCR 图片第 {attempt} 次失败 ({type(e).__name__}: {e})，{delay}s 后重试...\n{tb}")
+                    console.print(f"[yellow]OCR 图片第 {attempt} 次失败 ({type(e).__name__}: {e})，{delay}s 后重试...[/yellow]")
                     time.sleep(delay)
                 else:
-                    logger.error(f"OCR 图片全部 {max_retries} 次重试失败: {last_error}")
-                    console.print(f"[red]OCR 图片解析失败（已重试 {max_retries} 次）: {last_error}[/red]")
+                    logger.error(f"OCR 图片全部 {max_retries} 次重试失败 ({type(last_error).__name__}: {last_error})\n{tb}")
+                    console.print(f"[red]OCR 图片解析失败（已重试 {max_retries} 次）: {type(last_error).__name__}: {last_error}[/red]")
         if img_results:
             ocr_results.extend(img_results)
 
@@ -150,28 +157,40 @@ def _build_overlapping_batches(
     batch_size: int = 2,
     overlap: int = 1,
 ) -> List[List[Dict[str, Any]]]:
-    """构建重叠批次
+    """构建重叠批次，并为每页打上 is_primary 标记。
 
     每批 batch_size 页，相邻批次重叠 overlap 页。
     例如 5 页, batch_size=2, overlap=1:
-        批次0: [page0, page1]
-        批次1: [page1, page2]
-        批次2: [page2, page3]
-        批次3: [page3, page4]
+        批次0: [page0(primary), page1(context)]
+        批次1: [page1(primary), page2(context)]
+        批次2: [page2(primary), page3(context)]
+        批次3: [page3(primary), page4(primary←最后一批)]
+
+    is_primary=True  → Agent 必须提取该页开始的所有题目
+    is_primary=False → 仅作跨页上下文，Agent 不提取该页独立开始的题目
+
+    最后一批次的所有页均为 primary（最后一页没有后续批次来处理它）。
     """
     if not ocr_data:
         return []
 
     n_pages = len(ocr_data)
     if n_pages <= batch_size:
-        return [ocr_data]
+        # 只有一批：全部页都是 primary
+        return [[dict(page, is_primary=True) for page in ocr_data]]
 
     step = batch_size - overlap
     batches = []
     for start in range(0, n_pages, step):
         end = min(start + batch_size, n_pages)
-        batches.append(ocr_data[start:end])
-        if end >= n_pages:
+        is_last = (end >= n_pages)
+        batch = []
+        for page_idx, page in enumerate(ocr_data[start:end]):
+            # 第一页始终 primary；最后批次的所有页都 primary
+            primary = (page_idx == 0) or is_last
+            batch.append(dict(page, is_primary=primary))
+        batches.append(batch)
+        if is_last:
             break
 
     return batches
@@ -240,7 +259,7 @@ def _identify_subject(
 
     # ── 第 1 层：LLM 预检 ──
     try:
-        from error_correction_agent.agent import detect_subject_via_llm
+        from agents.error_correction.agent import detect_subject_via_llm
 
         llm_result = detect_subject_via_llm(text_sample, db_subjects, provider=model_provider)
         if llm_result:
@@ -299,18 +318,218 @@ def _content_fingerprint(q: Dict[str, Any]) -> str:
     return ""
 
 
+def _question_text(q: Dict[str, Any], chars: int = 400) -> str:
+    """提取题目纯文本，用于相似度计算。
+
+    包含：content_blocks 中的 text 块（去除 HTML 标签）+ options 选项列表。
+    image 块不参与计算。
+
+    chars 上限设为 400：覆盖大多数题目的完整内容，同时避免超长题目的 O(n²) 开销。
+
+    去除 HTML 标签的原因：表格题目在不同批次的 table 结构可能不同（th/td 排列差异），
+    但纯文本内容相同。去标签后两个版本的实验描述文字完全一致，相似度可正确判断。
+
+    纳入 options 的原因：跨页截断时题干可能不同，但选项往往相同，是最稳定的指纹。
+    """
+    parts = []
+    for block in q.get("content_blocks", []):
+        if block.get("block_type") == "text":
+            content = block.get("content", "")
+            content = _HTML_TAG_RE.sub("", content)   # 去除 HTML 标签，保留文本
+            parts.append(content)
+    for opt in q.get("options") or []:
+        if isinstance(opt, str):
+            parts.append(opt)
+    return "".join(parts)[:chars]
+
+
+def _image_paths(q: Dict[str, Any]) -> set:
+    """提取题目所有 image block 的路径集合，用于图片指纹比对。"""
+    paths = set()
+    for block in q.get("content_blocks", []):
+        if block.get("block_type") == "image":
+            p = block.get("content", "").strip()
+            if p:
+                paths.add(p)
+    return paths
+
+
+def _text_similarity(q1: Dict[str, Any], q2: Dict[str, Any]) -> float:
+    """计算两道题目内容相似度，返回 0~1。
+
+    优先用文本相似度（difflib）；若两题文本均过短（图片题），
+    则用图片路径 Jaccard 系数兜底，避免纯图片题的重复无法被检测到。
+    """
+    t1 = _question_text(q1)
+    t2 = _question_text(q2)
+
+    # 文本足够长时直接用 difflib
+    if len(t1) >= 10 and len(t2) >= 10:
+        return _difflib.SequenceMatcher(None, t1, t2).ratio()
+
+    # 文本过短（图片题）：用图片路径 Jaccard 系数兜底
+    imgs1 = _image_paths(q1)
+    imgs2 = _image_paths(q2)
+    if imgs1 or imgs2:
+        intersection = len(imgs1 & imgs2)
+        union = len(imgs1 | imgs2)
+        return intersection / union if union else 0.0
+
+    # 文本和图片都为空：不认为是重复
+    return 0.0
+
+
+def _fix_leading_images(questions: List[Dict[str, Any]]) -> None:
+    """后处理：将题目 content_blocks 中排在所有文本之前的图片移到前一道题末尾。
+
+    产生原因：试卷某题的插图印在下一页顶部，OCR 按页面顺序扫描时，该图片
+    出现在下一题文字之前。Agent 在该批次里将其归入下一题的 content_blocks 首位，
+    实际上属于上一题。
+
+    判断规则：若一道题的 content_blocks 中第一个出现的 text block 之前存在
+    image block，则这些 leading image 属于上一道题，移至上一题末尾。
+    """
+    for i in range(1, len(questions)):
+        q = questions[i]
+        blocks = q.get("content_blocks") or []
+        if not blocks:
+            continue
+
+        # 分离开头的连续 image 块（在第一个 text 块之前）
+        leading_images: List[Dict[str, Any]] = []
+        rest: List[Dict[str, Any]] = []
+        found_text = False
+        for block in blocks:
+            if not found_text and block.get("block_type") == "image":
+                leading_images.append(block)
+            else:
+                if block.get("block_type") == "text":
+                    found_text = True
+                rest.append(block)
+
+        if not leading_images:
+            continue
+
+        # 若去掉 leading images 后该题没有任何文本，说明它本身是纯图片题，不搬移
+        if not any(b.get("block_type") == "text" for b in rest):
+            continue
+
+        # 移到前一道题末尾
+        prev_q = questions[i - 1]
+        prev_blocks = prev_q.get("content_blocks") or []
+        prev_q["content_blocks"] = prev_blocks + leading_images
+        q["content_blocks"] = rest
+        logger.info(
+            f"修复 leading image: {len(leading_images)} 张图片 "
+            f"从题目 {q.get('question_id')} 移至题目 {prev_q.get('question_id')}"
+        )
+
+
+def _normalize_image_paths(questions: List[Dict[str, Any]]) -> None:
+    """修复 LLM 可能篡改的图片路径，原地修改。
+
+    LLM 有时会把输入中的 '/images/xxx.jpg' 改写为 'imgs/xxx.jpg'，
+    导致前端通过 Vite proxy 请求时路径不匹配。统一规范为 '/images/xxx'。
+    """
+    def _fix(path: str) -> str:
+        p = path.strip()
+        if p.startswith("imgs/"):
+            return "/images/" + p[len("imgs/"):]
+        return p
+
+    for q in questions:
+        for block in q.get("content_blocks") or []:
+            content = block.get("content", "")
+            if block.get("block_type") == "image":
+                block["content"] = _fix(content)
+            elif block.get("block_type") == "text" and "imgs/" in content:
+                # 修复 text block 中嵌入的 HTML 图片路径（如 table 中的 <img>）
+                block["content"] = content.replace('src="imgs/', 'src="/images/')
+        if q.get("image_refs"):
+            q["image_refs"] = [_fix(ref) for ref in q["image_refs"]]
+        if q.get("option_images"):
+            q["option_images"] = [_fix(ref) if ref else ref for ref in q["option_images"]]
+
+
+def _propagate_section_between_batches(batch_results: List[List[Dict[str, Any]]]) -> None:
+    """跨批次传播 section_title，原地修改 batch_results。
+
+    并行批次处理完成后调用。若批次 i 开头若干题的 section_title=None，
+    说明大题标题在前一批次的页面里，当前批次窗口看不到。
+    从批次 i-1 的末尾向前找最后一个有 section_title 的题，将其 section_title
+    赋给批次 i 中连续的 section=None 开头题，遇到批次 i 自己识别出 section_title
+    的题时停止传播。
+
+    防误传：若待传播 section 是在批次 i-1 中由题号较大的题建立的，则不向题号
+    更小的题传播（说明该 None 题属于更早的大题，而非当前 last_section）。
+    """
+    def _to_int_id(q) -> int | None:
+        try:
+            return int(str(q.get("question_id", "")).strip())
+        except (ValueError, TypeError):
+            return None
+
+    for i in range(1, len(batch_results)):
+        # 单次遍历上一批次：同时找 last_section 和该节首次出现的最小题号
+        last_section = None
+        first_section_id: int | None = None
+        for q in batch_results[i - 1]:
+            s = q.get("section_title")
+            if s:
+                if s != last_section:
+                    # 进入新节，重置首题号
+                    last_section = s
+                    first_section_id = None
+                qid = _to_int_id(q)
+                if qid is not None:
+                    if first_section_id is None or qid < first_section_id:
+                        first_section_id = qid
+
+        if not last_section:
+            continue
+
+        # 将 last_section 赋给本批次开头连续的 section=None 题
+        # 防误传：若待传播题的题号 < first_section_id，说明它属于更早的大题，跳过
+        propagated = 0
+        for q in batch_results[i]:
+            if not q.get("section_title"):
+                qid = _to_int_id(q)
+                if first_section_id is not None and qid is not None and qid < first_section_id:
+                    # 该题题号比 last_section 第一题还小，不应继承该 section
+                    continue
+                q["section_title"] = last_section
+                propagated += 1
+            else:
+                break  # 本批次已有自己识别出的 section，停止传播
+
+        if propagated:
+            logger.info(
+                f"跨批次 section 传播: 批次 {i} 前 {propagated} 道题 "
+                f"继承 section_title={repr(last_section)}"
+            )
+
+
 def _dedup_questions(questions: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
-    """按「题号 + 内容指纹」去重。
+    """按「(大题, 题号)」去重，再用内容相似度消除跨批次重复。
 
-    - 同题号 + 相同指纹 → 重叠页产生的真重复，保留内容更丰富的版本
-    - 同题号 + 不同指纹 → 不同板块的同号题（如小学试卷各大题下均有第1题），全部保留
+    第一轮：同一 (section_title, question_id) 内保留最丰富版本。
+    第二轮：同一 question_id 下若有多份，两两计算前 120 字相似度；
+            相似度 ≥ 0.75 视为重复，优先保留有 section_title 的版本，
+            相似度 < 0.75 视为不同大题下的同号题，全部保留。
     """
     if not questions:
         return []
 
     from collections import defaultdict
 
-    groups: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    # ── 预处理：将 question_id 统一归一化为 str，防止 int/str 类型不一致导致去重失效 ──
+    for q in questions:
+        qid = q.get("question_id")
+        if qid is not None:
+            q["question_id"] = str(qid).strip()
+
+    # ── 第一轮：按 (section, qid) 复合键去重 ──────────────────
+    groups: Dict[tuple, List[Dict[str, Any]]] = defaultdict(list)
     no_id: List[Dict[str, Any]] = []
 
     for q in questions:
@@ -318,26 +537,91 @@ def _dedup_questions(questions: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
         if not qid:
             no_id.append(q)
         else:
-            groups[qid].append(q)
-
-    result: List[Dict[str, Any]] = list(no_id)
+            section = q.get("section_title") or ""
+            groups[(section, qid)].append(q)
 
+    after_round1: List[Dict[str, Any]] = list(no_id)
     for qs in groups.values():
-        if len(qs) == 1:
-            result.append(qs[0])
+        after_round1.append(max(qs, key=_question_richness))
+
+    # ── 第二轮：结构优先 + 内容相似度去重 ───────────────────────
+    # 策略：
+    #   1. 同一 qid 下同时存在有 section 和无 section 版本 → 直接丢弃所有 section=None 版本
+    #      （section=None 说明该批次只捕获到选项/部分内容，有 section 的是完整版本）
+    #   2. 同一 qid 下均有 section（不同 section）→ difflib 相似度 ≥ 0.75 视为重复，保留最优
+    #   3. 同一 qid 下均无 section → 保留内容最丰富的一份
+    by_qid: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    for q in after_round1:
+        by_qid[q.get("question_id", "")].append(q)
+
+    SIMILARITY_THRESHOLD = 0.75
+    final: List[Dict[str, Any]] = list(no_id)
+    round2_removed = 0
+
+    for qid, entries in by_qid.items():
+        if not qid:
             continue
+        if len(entries) == 1:
+            final.extend(entries)
+            continue
+
+        sectioned = [q for q in entries if q.get("section_title")]
+        unsectioned = [q for q in entries if not q.get("section_title")]
 
-        # 按内容指纹二次分组
-        fp_groups: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
-        for q in qs:
-            fp_groups[_content_fingerprint(q)].append(q)
+        # 策略1：有 section 版本存在时，丢弃全部 section=None 版本
+        if sectioned and unsectioned:
+            round2_removed += len(unsectioned)
+            for q in unsectioned:
+                logger.debug(
+                    f"二次去重剔除(结构): qid={qid} "
+                    f"section=None，保留有大题标题的版本，richness={_question_richness(q)}"
+                )
+            entries = sectioned
+
+        if len(entries) == 1:
+            final.extend(entries)
+            continue
 
-        # 每个指纹组只保留最丰富的版本
-        for fp_qs in fp_groups.values():
-            result.append(max(fp_qs, key=_question_richness))
+        # 策略2/3：全部有 section 或全部无 section → difflib 去重
+        entries.sort(key=lambda q: (
+            0 if q.get("section_title") else 1,
+            -_question_richness(q)
+        ))
+
+        kept: List[Dict[str, Any]] = []
+        for candidate in entries:
+            sim = max(
+                (_text_similarity(candidate, k) for k in kept),
+                default=0.0,
+            )
+            if sim >= SIMILARITY_THRESHOLD:
+                round2_removed += 1
+                logger.debug(
+                    f"二次去重剔除(相似度): qid={qid} "
+                    f"section={repr(candidate.get('section_title'))} "
+                    f"相似度={sim:.2f}"
+                )
+            else:
+                kept.append(candidate)
+        final.extend(kept)
+
+    if round2_removed:
+        logger.info(f"二次去重: 剔除 {round2_removed} 道重复题目（结构优先 + 相似度阈值={SIMILARITY_THRESHOLD}）")
+        console.print(f"[yellow]二次去重: 剔除 {round2_removed} 道重复题目[/yellow]")
+
+    # ── 排序：有 section 的题按首次出现顺序 + 题号，section=None 的题排最后 ──
+    section_order: Dict[str, int] = {}
+    for q in questions:
+        s = q.get("section_title")
+        if s and s not in section_order:
+            section_order[s] = len(section_order)
 
-    result.sort(key=lambda q: _sort_key(q.get("question_id", "")))
-    return result
+    final.sort(key=lambda q: (
+        0 if q.get("section_title") else 1,          # 有 section 的排前，None 排后
+        section_order.get(q.get("section_title") or "", 999),
+        _sort_key(q.get("question_id", ""))
+    ))
+    return final
 
 
 def _question_richness(q: Dict[str, Any]) -> int:
@@ -388,30 +672,43 @@ def split_questions_node(state: WorkflowState) -> dict:
     if os.path.exists(meta_path):
         os.remove(meta_path)
 
-    # ── Step 1: OCR 解析 ──
-    console.print(f"[cyan]OCR 解析 {len(file_paths)} 个文件...[/cyan]")
-    ocr_start = time.time()
-
-    ocr_data = _run_ocr_and_simplify(file_paths, ocr_credentials=ocr_credentials)
+    # ── Step 1: OCR 解析（优先使用缓存） ──
+    ocr_cache_path = os.path.join(results_dir, "ocr_cache.json")
+    ocr_data = None
+    if os.path.exists(ocr_cache_path):
+        try:
+            with open(ocr_cache_path, 'r', encoding='utf-8') as f:
+                ocr_data = json.load(f)
+            console.print(f"[green]✓ 使用 OCR 缓存: {len(ocr_data)} 页[/green]")
+            logger.info(f"使用 OCR 缓存: {len(ocr_data)} 页")
+            os.remove(ocr_cache_path)  # 用完即删，避免下次误用
+        except Exception as e:
+            logger.warning(f"读取 OCR 缓存失败: {e}，将重新执行 OCR")
+            ocr_data = None
 
     if not ocr_data:
-        logger.error("OCR 解析失败，无数据返回")
-        console.print("[red]⚠ OCR 解析失败[/red]")
-        return {
-            "questions": [],
-            "warnings": ["步骤 2（OCR 解析）失败：无法解析图片内容，请检查 PaddleOCR API Token 配置"],
-        }
+        console.print(f"[cyan]OCR 解析 {len(file_paths)} 个文件...[/cyan]")
+        ocr_start = time.time()
+        ocr_data = _run_ocr_and_simplify(file_paths, ocr_credentials=ocr_credentials)
+
+        if not ocr_data:
+            logger.error("OCR 解析失败，无数据返回")
+            console.print("[red]⚠ OCR 解析失败[/red]")
+            return {
+                "questions": [],
+                "warnings": ["步骤 2（OCR 解析）失败：无法解析图片内容，请检查 PaddleOCR API Token 配置"],
+            }
+
+        total_blocks = sum(len(p.get("blocks", [])) for p in ocr_data)
+        ocr_elapsed = time.time() - ocr_start
+        logger.info(f"OCR 完成: {len(ocr_data)} 页, {total_blocks} 个 block, 耗时 {ocr_elapsed:.2f}s")
+        console.print(f"[green]✓ OCR 完成: {len(ocr_data)} 页, {total_blocks} 个 block ({ocr_elapsed:.1f}s)[/green]")
 
     # 保存 agent_input.json（供纠错节点使用）
     agent_input_path = os.path.join(results_dir, "agent_input.json")
     with open(agent_input_path, 'w', encoding='utf-8') as f:
         json.dump(ocr_data, f, ensure_ascii=False, indent=2)
 
-    total_blocks = sum(len(p.get("blocks", [])) for p in ocr_data)
-    ocr_elapsed = time.time() - ocr_start
-    logger.info(f"OCR 完成: {len(ocr_data)} 页, {total_blocks} 个 block, 耗时 {ocr_elapsed:.2f}s")
-    console.print(f"[green]✓ OCR 完成: {len(ocr_data)} 页, {total_blocks} 个 block ({ocr_elapsed:.1f}s)[/green]")
-
     # ── Step 2: 加载 DB 上下文 ──
     db_subjects, db_tags = _load_db_context()
 
@@ -435,13 +732,18 @@ def split_questions_node(state: WorkflowState) -> dict:
 
     # ── Step 4: 构建重叠批次 ──
     batches = _build_overlapping_batches(ocr_data, batch_size=2, overlap=1)
-    console.print(f"[cyan]构建 {len(batches)} 个批次（2页/批, 1页重叠）[/cyan]")
+    console.print(f"[cyan]构建 {len(batches)} 个批次（每批1主页+1上下文页）[/cyan]")
+
+    # 保存完整批次数据到 agent_input.json（含 is_primary 标记，供调试和纠错节点使用）
+    agent_input_path = os.path.join(results_dir, "agent_input.json")
+    with open(agent_input_path, 'w', encoding='utf-8') as f:
+        json.dump(batches, f, ensure_ascii=False, indent=2)
 
     # ── Step 5: 并行分割 ──
     split_start = time.time()
     console.print(f"[cyan]并行分割 {len(batches)} 个批次...[/cyan]")
 
-    from error_correction_agent.tools import split_batch
+    from agents.error_correction.tools import split_batch
 
     existing_tags_str = ",".join(db_tags) if db_tags else ""
     max_workers = min(len(batches), 3)
@@ -497,7 +799,9 @@ def _invoke_split(batch_idx: int, batch_data: list) -> None:
     split_elapsed = time.time() - split_start
     logger.info(f"并行分割完成, 耗时 {split_elapsed:.2f}s")
 
-    # ── Step 6: 合并 + 去重 ──
+    # ── Step 6: 跨批次 section 传播 + 合并 + 去重 ──
+    _propagate_section_between_batches(batch_results)
+
     all_questions = []
     for questions in batch_results:
         all_questions.extend(questions)
@@ -510,6 +814,16 @@ def _invoke_split(batch_idx: int, batch_data: list) -> None:
         logger.info(f"去重: {before_dedup} → {after_dedup} 道题目（移除 {before_dedup - after_dedup} 道重复）")
         console.print(f"[yellow]去重: 移除 {before_dedup - after_dedup} 道重复题目[/yellow]")
 
+    # 修复 leading image（页面顶部图属于上一题）
+    _fix_leading_images(deduped)
+
+    # 修复 LLM 可能篡改的图片路径（如 imgs/xxx → /images/xxx）
+    _normalize_image_paths(deduped)
+
+    # 为每道题赋全局唯一 uid（顺序字符串），供前端选择和导出时使用
+    for i, q in enumerate(deduped):
+        q["uid"] = str(i)
+
     # ── Step 7: 保存结果 ──
     with open(questions_file, 'w', encoding='utf-8') as f:
         json.dump(deduped, f, ensure_ascii=False, indent=2)
@@ -567,15 +881,29 @@ def correct_questions_node(state: WorkflowState) -> dict:
     console.print(f"[cyan]发现 {len(flagged)} 道题目需要纠错（共 {len(questions)} 道）[/cyan]")
     logger.info(f"开始纠错: {len(flagged)}/{len(questions)} 道题目")
 
-    # 加载原始 OCR 数据作为纠错上下文
+    # 加载 OCR 数据作为纠错上下文
+    # agent_input.json 现在保存的是批次数据（[[page, ...], ...]），需要还原为扁平页面列表
     ocr_context = "{}"
     agent_input_path = os.path.join(settings.results_dir, "agent_input.json")
     if os.path.exists(agent_input_path):
         with open(agent_input_path, 'r', encoding='utf-8') as f:
-            ocr_context = f.read()
+            raw = json.load(f)
+        # 兼容新格式（批次列表）和旧格式（扁平页面列表）
+        if raw and isinstance(raw[0], list):
+            # 新格式：按 page_index 去重，还原为扁平页面列表
+            seen = {}
+            for batch in raw:
+                for page in batch:
+                    idx = page.get("page_index", id(page))
+                    if idx not in seen:
+                        seen[idx] = page
+            flat_pages = [seen[k] for k in sorted(seen)]
+            ocr_context = json.dumps(flat_pages, ensure_ascii=False)
+        else:
+            ocr_context = json.dumps(raw, ensure_ascii=False)
 
     # 调用纠错工具
-    from error_correction_agent.tools import correct_batch
+    from agents.error_correction.tools import correct_batch
 
     model_provider = state.get("model_provider", "openai")
     model_name = state.get("model_name")
@@ -597,19 +925,25 @@ def correct_questions_node(state: WorkflowState) -> dict:
         console.print("[red]⚠ 纠错结果解析失败，保留原始题目[/red]")
         return {"questions": questions}
 
-    # 按 question_id 合并纠错结果
-    corrected_map = {q["question_id"]: q for q in corrected}
+    # 按 (section_title, question_id) 复合键合并纠错结果，避免不同大题同号题互相覆盖
+    corrected_map = {
+        (q.get("section_title") or "", q["question_id"]): q
+        for q in corrected
+    }
 
     merged = []
     for q in questions:
         qid = q.get("question_id")
-        if qid in corrected_map:
-            cq = corrected_map[qid]
+        section = q.get("section_title") or ""
+        key = (section, qid)
+        if key in corrected_map:
+            cq = corrected_map[key]
             corrections = cq.pop("corrections_applied", [])
             cq["needs_correction"] = False
             cq["ocr_issues"] = None
+            cq["uid"] = q.get("uid")  # 纠错对象来自 LLM 输出，不含 uid，从原题还原
             merged.append(cq)
-            logger.info(f"题目 {qid} 已纠错: {corrections}")
+            logger.info(f"题目 {section}-{qid} 已纠错: {corrections}")
         else:
             merged.append(q)
 
diff --git a/backend/tests/test_agent_capabilities.py b/backend/tests/test_agent_capabilities.py
new file mode 100644
index 00000000..1a051b3f
--- /dev/null
+++ b/backend/tests/test_agent_capabilities.py
@@ -0,0 +1,21 @@
+from agents.error_correction.agent import _should_avoid_tool_strategy
+
+
+class _DummyCfg:
+    def __init__(self, model_name: str, base_url: str = ""):
+        self._model_name = model_name
+        self.base_url = base_url
+
+    def resolve_model_name(self, model_name=None, *, use_light=False):
+        return model_name or self._model_name
+
+
+def test_should_avoid_tool_strategy_for_deepseek_reasoner():
+    cfg = _DummyCfg(model_name="deepseek-reasoner", base_url="https://api.deepseek.com")
+    assert _should_avoid_tool_strategy("openai", None, cfg) is True
+
+
+def test_should_not_avoid_tool_strategy_for_deepseek_chat():
+    cfg = _DummyCfg(model_name="deepseek-chat", base_url="https://api.deepseek.com")
+    assert _should_avoid_tool_strategy("openai", None, cfg) is False
+
diff --git a/backend/tests/test_chat_routes.py b/backend/tests/test_chat_routes.py
index 4b856bcc..76e16ba0 100644
--- a/backend/tests/test_chat_routes.py
+++ b/backend/tests/test_chat_routes.py
@@ -52,10 +52,17 @@ def __enter__(self):
         def __exit__(self, *args):
             pass
 
-    with patch("web_app.SessionLocal", FakeSessionLocal()):
+    fake = FakeSessionLocal()
+    with patch("routes.chat.SessionLocal", fake), \
+         patch("routes.questions.SessionLocal", fake), \
+         patch("routes.settings.SessionLocal", fake), \
+         patch("routes.auth.SessionLocal", fake):
         from web_app import app
         app.config["TESTING"] = True
         with app.test_client() as c:
+            with c.session_transaction() as sess:
+                sess['user_id'] = 1
+                sess['username'] = 'test'
             yield c
 
 
diff --git a/backend/tests/test_correct_node.py b/backend/tests/test_correct_node.py
index 5aa58c1c..3a6c63c6 100644
--- a/backend/tests/test_correct_node.py
+++ b/backend/tests/test_correct_node.py
@@ -40,7 +40,7 @@ def test_skip_when_no_flagged(self):
         result = correct_questions_node(state)
         assert result["questions"] == qs
 
-    @patch("error_correction_agent.tools.correct_batch")
+    @patch("agents.error_correction.tools.correct_batch")
     def test_merge_corrected(self, mock_correct_batch, tmp_path):
         """纠错结果应按 question_id 合并回原列表"""
         q1 = _q("1")
@@ -62,7 +62,7 @@ def test_merge_corrected(self, mock_correct_batch, tmp_path):
         with open(os.path.join(results_dir, "agent_input.json"), "w") as f:
             f.write(agent_input)
 
-        with patch("config.settings.results_dir", Path(results_dir)):
+        with patch("core.config.settings.results_dir", Path(results_dir)):
             state = {"questions": [q1, q2, q3]}
             result = correct_questions_node(state)
 
@@ -79,7 +79,7 @@ def test_merge_corrected(self, mock_correct_batch, tmp_path):
         merged_q1 = next(q for q in merged if q["question_id"] == "1")
         assert merged_q1["content_blocks"][0]["content"] == "内容"
 
-    @patch("error_correction_agent.tools.correct_batch")
+    @patch("agents.error_correction.tools.correct_batch")
     def test_invalid_json_keeps_original(self, mock_correct_batch, tmp_path):
         """纠错返回无效 JSON 时应保留原始题目"""
         q1 = _q("1", needs_correction=True)
@@ -89,7 +89,7 @@ def test_invalid_json_keeps_original(self, mock_correct_batch, tmp_path):
         with open(os.path.join(results_dir, "agent_input.json"), "w") as f:
             f.write("{}")
 
-        with patch("config.settings.results_dir", Path(results_dir)):
+        with patch("core.config.settings.results_dir", Path(results_dir)):
             state = {"questions": [q1]}
             result = correct_questions_node(state)
 
diff --git a/backend/tests/test_ocr_api.py b/backend/tests/test_ocr_api.py
index 14cf78e6..57474a30 100644
--- a/backend/tests/test_ocr_api.py
+++ b/backend/tests/test_ocr_api.py
@@ -5,7 +5,7 @@
     pytest tests/test_ocr_api.py -v -s
 
 需要配置环境变量：PADDLEOCR_API_URL、PADDLEOCR_API_TOKEN。
-测试图片使用 example_uploads/ 目录下的文件（test.jpg + test4.pdf）。
+测试图片使用 example_uploads/notes/test.jpg，测试 PDF 使用 example_uploads/exams/test4.pdf。
 """
 
 import os
@@ -19,15 +19,46 @@
 
 # ── 配置 ────────────────────────────────────────────────
 
-API_URL = os.getenv("PADDLEOCR_API_URL", "")
-TOKEN = os.getenv("PADDLEOCR_API_TOKEN", "")
-MODEL = os.getenv("PADDLEOCR_MODEL", "PaddleOCR-VL-1.5")
+
+def _load_ocr_creds_from_db() -> dict:
+    """从数据库读取激活的 PaddleOCR provider 凭据（作为环境变量的 fallback）"""
+    try:
+        import sys
+        _backend = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+        if _backend not in sys.path:
+            sys.path.insert(0, _backend)
+        from db import SessionLocal
+        from db.models import ProviderConfig
+        with SessionLocal() as db:
+            p = db.query(ProviderConfig).filter(
+                ProviderConfig.category == "paddleocr",
+                ProviderConfig.is_active == True,
+            ).first()
+            if p:
+                return {
+                    "api_url": p.base_url or "",
+                    "token": p.api_key or "",
+                    "model": p.model_name or "PaddleOCR-VL-1.5",
+                    "use_doc_orientation": p.use_doc_orientation,
+                    "use_doc_unwarping": p.use_doc_unwarping,
+                    "use_chart_recognition": p.use_chart_recognition,
+                }
+    except Exception:
+        pass
+    return {}
+
+
+_DB_CREDS = _load_ocr_creds_from_db()
+
+API_URL = os.getenv("PADDLEOCR_API_URL") or _DB_CREDS.get("api_url", "")
+TOKEN = os.getenv("PADDLEOCR_API_TOKEN") or _DB_CREDS.get("token", "")
+MODEL = os.getenv("PADDLEOCR_MODEL") or _DB_CREDS.get("model", "PaddleOCR-VL-1.5")
 
 EXAMPLE_DIR = os.path.abspath(
     os.path.join(os.path.dirname(__file__), "..", "..", "example_uploads")
 )
-TEST_IMAGE = os.path.join(EXAMPLE_DIR, "test.jpg")
-TEST_PDF = os.path.join(EXAMPLE_DIR, "test4.pdf")
+TEST_IMAGE = os.path.join(EXAMPLE_DIR, "notes", "test.jpg")
+TEST_PDF = os.path.join(EXAMPLE_DIR, "exams", "test4.pdf")
 
 POLL_INTERVAL = 3
 POLL_TIMEOUT = 120  # 最长等待秒数
@@ -144,7 +175,7 @@ class TestImageApiConnection:
     def test_submit_returns_job_id(self, image_job_result):
         """提交图片任务应返回非空 jobId"""
         job_id, _ = image_job_result
-        assert job_id and job_id.startswith("ocrjob-")
+        assert job_id
 
     def test_job_completes_successfully(self, image_job_result):
         """图片任务应在超时前完成"""
@@ -242,7 +273,7 @@ class TestPdfApiConnection:
     def test_submit_returns_job_id(self, pdf_job_result):
         """提交 PDF 任务应返回非空 jobId"""
         job_id, _ = pdf_job_result
-        assert job_id and job_id.startswith("ocrjob-")
+        assert job_id
 
     def test_job_completes_successfully(self, pdf_job_result):
         """PDF 任务应在超时前完成"""
@@ -295,7 +326,7 @@ def test_parse_image_returns_result(self, tmp_path):
         """parse_image 应返回包含 layoutParsingResults 的结果"""
         from src.paddleocr_client import PaddleOCRClient
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         result = client.parse_image(TEST_IMAGE, save_output=True, output_dir=str(tmp_path))
 
         assert "layoutParsingResults" in result
@@ -306,7 +337,7 @@ def test_saves_struct_json(self, tmp_path):
         from src.paddleocr_client import PaddleOCRClient
         from pathlib import Path
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         client.parse_image(TEST_IMAGE, save_output=True, output_dir=str(tmp_path))
 
         stem = Path(TEST_IMAGE).stem
@@ -322,7 +353,7 @@ def test_simplify_ocr_results_compatible(self):
         from src.paddleocr_client import PaddleOCRClient
         from src.utils import simplify_ocr_results
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         result = client.parse_image(TEST_IMAGE, save_output=False)
 
         simplified = simplify_ocr_results([result])
@@ -342,7 +373,7 @@ def test_parse_pdf_returns_result(self, tmp_path):
         """parse_pdf 应返回包含 layoutParsingResults 的结果"""
         from src.paddleocr_client import PaddleOCRClient
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         result = client.parse_pdf(TEST_PDF, save_output=True, output_dir=str(tmp_path))
 
         assert "layoutParsingResults" in result
@@ -353,7 +384,7 @@ def test_pdf_simplify_compatible(self):
         from src.paddleocr_client import PaddleOCRClient
         from src.utils import simplify_ocr_results
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         result = client.parse_pdf(TEST_PDF, save_output=False)
 
         simplified = simplify_ocr_results([result])
@@ -367,7 +398,7 @@ def test_pdf_multipage_index_continuity(self):
         from src.paddleocr_client import PaddleOCRClient
         from src.utils import simplify_ocr_results
 
-        client = PaddleOCRClient()
+        client = PaddleOCRClient(**_DB_CREDS)
         result = client.parse_pdf(TEST_PDF, save_output=False)
 
         simplified = simplify_ocr_results([result])
diff --git a/backend/tests/test_question_tools.py b/backend/tests/test_question_tools.py
index ff262cc1..f1af1cf5 100644
--- a/backend/tests/test_question_tools.py
+++ b/backend/tests/test_question_tools.py
@@ -10,7 +10,7 @@
 import json
 import pytest
 from unittest.mock import patch
-from error_correction_agent.tools.question_tools import save_questions, log_issue
+from agents.error_correction.tools.question_tools import save_questions, log_issue
 
 
 # ═══════════════════════════════════════════════════════════
@@ -25,7 +25,7 @@ def test_save_new_file(self, tmp_path):
         out = str(tmp_path / "questions.json")
         questions = [{"question_id": "1", "content_blocks": []}]
 
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             result = save_questions.invoke({
                 "questions": questions,
                 "output_path": out,
@@ -41,7 +41,7 @@ def test_append_to_existing(self, tmp_path):
         q1 = [{"question_id": "1", "content_blocks": []}]
         q2 = [{"question_id": "2", "content_blocks": []}]
 
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             save_questions.invoke({"questions": q1, "output_path": out})
             save_questions.invoke({"questions": q2, "output_path": out})
 
@@ -53,7 +53,7 @@ def test_subject_metadata(self, tmp_path):
         out = str(tmp_path / "questions.json")
         questions = [{"question_id": "1", "content_blocks": []}]
 
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             save_questions.invoke({
                 "questions": questions,
                 "subject": "高中数学",
@@ -70,7 +70,7 @@ def test_empty_subject_no_metadata(self, tmp_path):
         out = str(tmp_path / "questions.json")
         questions = [{"question_id": "1", "content_blocks": []}]
 
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             save_questions.invoke({
                 "questions": questions,
                 "subject": "",
@@ -90,7 +90,7 @@ class TestLogIssue:
     """log_issue 工具测试"""
 
     def test_basic_log(self, tmp_path):
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             result = log_issue.invoke({
                 "issue_type": "unclear_boundary",
                 "description": "第3题和第4题边界不清",
@@ -105,7 +105,7 @@ def test_basic_log(self, tmp_path):
         assert "第3题" in line["description"]
 
     def test_with_block_info(self, tmp_path):
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             log_issue.invoke({
                 "issue_type": "missing_question_number",
                 "description": "缺少题号",
@@ -119,7 +119,7 @@ def test_with_block_info(self, tmp_path):
 
     def test_append_multiple(self, tmp_path):
         """多次调用应追加（JSONL 格式）"""
-        with patch("config.settings.results_dir", tmp_path):
+        with patch("core.config.settings.results_dir", tmp_path):
             log_issue.invoke({"issue_type": "a", "description": "issue 1"})
             log_issue.invoke({"issue_type": "b", "description": "issue 2"})
 
diff --git a/backend/tests/test_schemas.py b/backend/tests/test_schemas.py
index 178e1d7a..f89b2f27 100644
--- a/backend/tests/test_schemas.py
+++ b/backend/tests/test_schemas.py
@@ -11,7 +11,7 @@
 
 import pytest
 from pydantic import ValidationError
-from error_correction_agent.schemas import (
+from agents.error_correction.schemas import (
     ContentBlock,
     Question,
     QuestionSplitResult,
diff --git a/backend/tests/test_solve_integration.py b/backend/tests/test_solve_integration.py
index 8b533a79..00ab1643 100644
--- a/backend/tests/test_solve_integration.py
+++ b/backend/tests/test_solve_integration.py
@@ -55,7 +55,7 @@
 @pytest.fixture(scope="session")
 def solve_results(model_provider):
     """session 级 fixture：只调用一次 API，所有测试共享结果"""
-    from solve_agent import invoke_solve
+    from agents.solve import invoke_solve
     return invoke_solve(SAMPLE_QUESTIONS, provider=model_provider)
 
 
diff --git a/backend/tests/test_solve_schemas.py b/backend/tests/test_solve_schemas.py
index e887dcf8..b843049b 100644
--- a/backend/tests/test_solve_schemas.py
+++ b/backend/tests/test_solve_schemas.py
@@ -8,7 +8,7 @@
 
 import pytest
 from pydantic import ValidationError
-from solve_agent.schemas import SolveResult, SolveBatchResult
+from agents.solve.schemas import SolveResult, SolveBatchResult
 
 
 # ═══════════════════════════════════════════════════════════
diff --git a/backend/tests/test_split_integration.py b/backend/tests/test_split_integration.py
index 0e07b037..0c314ca4 100644
--- a/backend/tests/test_split_integration.py
+++ b/backend/tests/test_split_integration.py
@@ -48,7 +48,7 @@ def ocr_data():
 @pytest.fixture(scope="session")
 def split_result(ocr_data, model_provider):
     """调用 split_batch 一次，所有测试共享结果（节省 API 调用）"""
-    from error_correction_agent.tools import split_batch
+    from agents.error_correction.tools import split_batch
 
     result = split_batch.invoke({
         "ocr_data": json.dumps(ocr_data, ensure_ascii=False),
@@ -79,7 +79,7 @@ def test_returns_non_empty(self, split_result):
 
     def test_question_schema(self, split_result):
         """每道题目应符合 Question schema"""
-        from error_correction_agent.schemas import Question
+        from agents.error_correction.schemas import Question
 
         for q_data in split_result:
             q = Question(**q_data)
diff --git a/backend/tests/test_utils.py b/backend/tests/test_utils.py
index 3c8bd52f..e73a408a 100644
--- a/backend/tests/test_utils.py
+++ b/backend/tests/test_utils.py
@@ -14,7 +14,7 @@
 def mock_env(monkeypatch, tmp_path):
     """使用临时目录作为 settings.pages_dir"""
     pages_dir = tmp_path / "pages"
-    monkeypatch.setattr("config.settings.pages_dir", pages_dir)
+    monkeypatch.setattr("core.config.settings.pages_dir", pages_dir)
     return pages_dir
 
 
diff --git a/backend/tests/test_web_helpers.py b/backend/tests/test_web_helpers.py
index 9bdad5fe..1c9ced04 100644
--- a/backend/tests/test_web_helpers.py
+++ b/backend/tests/test_web_helpers.py
@@ -8,7 +8,8 @@
 
 import os
 import pytest
-from web_app import allowed_file, _safe_join
+from routes.upload import allowed_file
+from web_app import _safe_join
 
 
 # ═══════════════════════════════════════════════════════════
diff --git a/backend/tests/test_web_routes.py b/backend/tests/test_web_routes.py
index b7f1bac4..9e7e775f 100644
--- a/backend/tests/test_web_routes.py
+++ b/backend/tests/test_web_routes.py
@@ -18,13 +18,16 @@
 from sqlalchemy import create_engine, event
 from sqlalchemy.orm import sessionmaker
 
-from db.models import Base
+from db.models import Base, User
 from db import crud
 
+# 测试用户 ID，与 client fixture 中的 session['user_id'] 一致
+TEST_USER_ID = 1
+
 
 @pytest.fixture
 def test_db():
-    """内存数据库 + 建表"""
+    """内存数据库 + 建表 + 创建测试用户"""
     engine = create_engine("sqlite:///:memory:", echo=False)
 
     @event.listens_for(engine, "connect")
@@ -36,6 +39,10 @@ def set_sqlite_pragma(dbapi_connection, connection_record):
     Base.metadata.create_all(bind=engine)
     Session = sessionmaker(bind=engine)
     session = Session()
+    # 创建测试用户（与 client session 中的 user_id 匹配）
+    user = User(id=TEST_USER_ID, username="test", email="test@test.com", password_hash="x")
+    session.add(user)
+    session.commit()
     yield session
     session.close()
 
@@ -54,10 +61,20 @@ def __enter__(self):
         def __exit__(self, *args):
             pass
 
-    with patch("web_app.SessionLocal", FakeSessionLocal()):
+    fake = FakeSessionLocal()
+    with patch("routes.settings.SessionLocal", fake), \
+         patch("routes.questions.SessionLocal", fake), \
+         patch("routes.stats.SessionLocal", fake), \
+         patch("routes.upload.SessionLocal", fake), \
+         patch("routes.chat.SessionLocal", fake), \
+         patch("routes.auth.SessionLocal", fake):
         from web_app import app
         app.config["TESTING"] = True
         with app.test_client() as c:
+            with c.session_transaction() as sess:
+                sess['user_id'] = TEST_USER_ID
+                sess['username'] = 'test'
+                sess['session_version'] = 0
             yield c
 
 
@@ -176,7 +193,7 @@ def test_delete_existing(self, client, test_db):
         save_questions_to_db(test_db, qs, {
             "original_filename": "test.pdf",
             "subject": "数学",
-        })
+        }, user_id=TEST_USER_ID)
 
         # 查询刚插入的题目 ID
         from db.models import Question
@@ -230,7 +247,7 @@ def test_with_data(self, client, test_db):
         save_questions_to_db(test_db, qs, {
             "original_filename": "test.pdf",
             "subject": "数学",
-        })
+        }, user_id=TEST_USER_ID)
         resp = client.get("/api/error-bank")
         data = resp.get_json()
         assert data["total"] == 1
@@ -260,7 +277,7 @@ def test_with_data(self, client, test_db):
             "question_type": "选择题",
             "content_blocks": [{"block_type": "text", "content": "t"}],
             "has_formula": False, "has_image": False,
-        }], {"original_filename": "a.pdf", "subject": "高中数学"})
+        }], {"original_filename": "a.pdf", "subject": "高中数学"}, user_id=TEST_USER_ID)
         resp = client.get("/api/subjects")
         assert "高中数学" in resp.get_json()["subjects"]
 
@@ -300,7 +317,7 @@ def test_save_answer(self, client, test_db):
             "question_type": "选择题",
             "content_blocks": [{"block_type": "text", "content": "答案测试"}],
             "has_formula": False, "has_image": False,
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         from db.models import Question
         q = test_db.query(Question).first()
         resp = client.patch(f"/api/question/{q.id}/answer", json={"user_answer": "选B"})
@@ -342,7 +359,7 @@ def test_invalid_status(self, client, test_db):
             "question_type": "选择题",
             "content_blocks": [{"block_type": "text", "content": "状态测试"}],
             "has_formula": False, "has_image": False,
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         from db.models import Question
         q = test_db.query(Question).first()
         resp = client.patch(f"/api/question/{q.id}/review-status", json={"review_status": "无效"})
@@ -359,7 +376,7 @@ def test_update_status(self, client, test_db):
             "question_type": "选择题",
             "content_blocks": [{"block_type": "text", "content": "复习状态测试"}],
             "has_formula": False, "has_image": False,
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         from db.models import Question
         q = test_db.query(Question).first()
         resp = client.patch(f"/api/question/{q.id}/review-status", json={"review_status": "已掌握"})
@@ -396,7 +413,7 @@ def test_with_data(self, client, test_db):
             "content_blocks": [{"block_type": "text", "content": "统计测试"}],
             "has_formula": False, "has_image": False,
             "knowledge_tags": ["导数"],
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         resp = client.get("/api/dashboard-stats")
         data = resp.get_json()
         assert data["total_questions"] == 1
@@ -413,14 +430,14 @@ def test_subject_filter(self, client, test_db):
             "content_blocks": [{"block_type": "text", "content": "数学题"}],
             "has_formula": False, "has_image": False,
             "knowledge_tags": ["函数"],
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         save_questions_to_db(test_db, [{
             "question_id": "2",
             "question_type": "选择题",
             "content_blocks": [{"block_type": "text", "content": "英语题"}],
             "has_formula": False, "has_image": False,
             "knowledge_tags": ["语法"],
-        }], {"original_filename": "b.pdf", "subject": "英语"})
+        }], {"original_filename": "b.pdf", "subject": "英语"}, user_id=TEST_USER_ID)
         resp = client.get("/api/dashboard-stats")
         data = resp.get_json()
         assert data["total_questions"] == 2
@@ -451,7 +468,7 @@ def test_filter_by_review_status(self, client, test_db):
                 "content_blocks": [{"block_type": "text", "content": "已掌握题"}],
                 "has_formula": False, "has_image": False,
             },
-        ], {"original_filename": "a.pdf", "subject": "数学"})
+        ], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         from db.models import Question
         qs = test_db.query(Question).all()
         update_review_status(test_db, qs[1].id, "已掌握")
@@ -492,7 +509,7 @@ def test_success(self, client, test_db):
             "content_blocks": [{"block_type": "text", "content": "AI分析测试"}],
             "has_formula": False, "has_image": False,
             "knowledge_tags": ["导数"],
-        }], {"original_filename": "a.pdf", "subject": "数学"})
+        }], {"original_filename": "a.pdf", "subject": "数学"}, user_id=TEST_USER_ID)
         from db.models import Question
         q = test_db.query(Question).first()
         resp = client.post("/api/ai-analysis", json={"question_ids": [q.id]})
diff --git a/backend/tests/test_workflow_helpers.py b/backend/tests/test_workflow_helpers.py
index 57b648f1..27b948ed 100644
--- a/backend/tests/test_workflow_helpers.py
+++ b/backend/tests/test_workflow_helpers.py
@@ -519,7 +519,7 @@ class TestIdentifySubject:
     @pytest.fixture(autouse=True)
     def _mock_llm(self):
         with patch(
-            "error_correction_agent.agent.detect_subject_via_llm",
+            "agents.error_correction.agent.detect_subject_via_llm",
             return_value="",
         ) as mock:
             self.mock_llm = mock
@@ -655,7 +655,7 @@ def _make_ocr(self, text: str) -> list:
             }]
         }]
 
-    @patch("error_correction_agent.agent.detect_subject_via_llm", return_value="高中数学")
+    @patch("agents.error_correction.agent.detect_subject_via_llm", return_value="高中数学")
     def test_llm_success(self, mock_llm):
         """LLM 返回有效科目时直接采用"""
         ocr = self._make_ocr("普通内容无关键词")
@@ -663,14 +663,14 @@ def test_llm_success(self, mock_llm):
         assert result == "高中数学"
         mock_llm.assert_called_once()
 
-    @patch("error_correction_agent.agent.detect_subject_via_llm", return_value="初中地理")
+    @patch("agents.error_correction.agent.detect_subject_via_llm", return_value="初中地理")
     def test_llm_returns_new_subject(self, mock_llm):
         """LLM 返回不在 db_subjects 中的新科目也应采用"""
         ocr = self._make_ocr("普通内容")
         result = _identify_subject(ocr, ["高中数学"])
         assert result == "初中地理"
 
-    @patch("error_correction_agent.agent.detect_subject_via_llm", return_value="")
+    @patch("agents.error_correction.agent.detect_subject_via_llm", return_value="")
     def test_llm_empty_fallback_to_keyword(self, mock_llm):
         """LLM 返回空时应 fallback 到关键词匹配"""
         ocr = self._make_ocr("数学试卷")
@@ -678,7 +678,7 @@ def test_llm_empty_fallback_to_keyword(self, mock_llm):
         assert result == "高中数学"
 
     @patch(
-        "error_correction_agent.agent.detect_subject_via_llm",
+        "agents.error_correction.agent.detect_subject_via_llm",
         side_effect=Exception("API error"),
     )
     def test_llm_exception_fallback(self, mock_llm):
@@ -687,7 +687,7 @@ def test_llm_exception_fallback(self, mock_llm):
         result = _identify_subject(ocr, [])
         assert result == "高中物理"
 
-    @patch("error_correction_agent.agent.detect_subject_via_llm", return_value="高中物理")
+    @patch("agents.error_correction.agent.detect_subject_via_llm", return_value="高中物理")
     def test_llm_receives_provider(self, mock_llm):
         """model_provider 应正确传递到 LLM 函数"""
         ocr = self._make_ocr("普通内容")
diff --git a/backend/web_app.py b/backend/web_app.py
index 8411ea0c..6e576aba 100644
--- a/backend/web_app.py
+++ b/backend/web_app.py
@@ -1,56 +1,66 @@
 """
-错题本生成系统 - Web应用
-提供前端界面用于执行工作流
+错题本生成系统 - Web应用（纯 API 服务）
+
+职责：
+  - 创建 Flask 应用实例，配置全局参数
+  - 注册所有 Blueprint 路由（routes/ 目录）
+  - 提供全局错误处理和登录鉴权中间件
+  - 提供 OCR 图片、下载文件、擦除结果等静态资源接口
+
+前端由 Vite dev server 独立运行（localhost:5173），
+通过代理将 /api 等请求转发到本服务（localhost:5001）。
 """
 
 import os
-import json
-import uuid
+import sys
 import logging
-import threading
-import mimetypes
-from datetime import datetime
-from pathlib import PurePath
-from typing import Optional
 
-import requests as http_requests
+# 无论从项目根目录执行 `python backend/web_app.py` 还是在 `backend` 下执行 `python web_app.py`，
+# 都把 backend 目录加入 sys.path，保证 `core`、`routes`、`db` 等包解析一致。
+_BACKEND_ROOT = os.path.dirname(os.path.abspath(__file__))
+if _BACKEND_ROOT not in sys.path:
+    sys.path.insert(0, _BACKEND_ROOT)
 
-# Windows 注册表可能将 .js 映射为 text/plain，导致浏览器拒绝加载 ES module
-mimetypes.add_type('application/javascript', '.js')
-from flask import Flask, request, jsonify, send_file, send_from_directory, redirect, session
+from flask import Flask, request, jsonify, send_file, session
+from flask_cors import CORS
 from dotenv import load_dotenv
 
-from werkzeug.security import generate_password_hash, check_password_hash
-from src.workflow import build_workflow
-from src.utils import export_wrongbook as export_wrongbook_md
-from config import settings
+from core.config import settings
 from db import init_db, SessionLocal
 from db import crud
-from db.models import Question, UploadBatch, KnowledgeTag
+from routes import register_routes
 
-# 加载环境变量
-load_dotenv()
+# 加载 backend/.env（无论从哪个目录启动都指向同一文件）
+load_dotenv(os.path.join(_BACKEND_ROOT, ".env"))
 
+# 模块级日志记录器，日志名称为 'web_app'
 logger = logging.getLogger(__name__)
 
+# ============================================================
+# Flask 应用初始化
+# ============================================================
+
 app = Flask(__name__)
+
+# 会话密钥：用于加密 Flask session cookie，生产环境必须修改
 app.secret_key = os.getenv('SECRET_KEY', 'dev-secret-change-in-production')
 
-# 配置（统一从 config.py 导入）
+# 跨域支持：前端 dev server (5173) 和后端 (5001) 端口不同，需要 CORS
+# supports_credentials=True 允许携带 cookie（登录态依赖 session cookie）
+CORS(app, supports_credentials=True)
+
+# 文件上传配置
 app.config['UPLOAD_FOLDER'] = settings.upload_dir
-app.config['MAX_CONTENT_LENGTH'] = settings.max_file_size_mb * 1024 * 1024
+app.config['MAX_CONTENT_LENGTH'] = settings.max_file_size_mb * 1024 * 1024  # MB → 字节
 
 
-def _safe_join(base_dir: str, rel_path: str) -> str | None:
-    base = os.path.abspath(base_dir)
-    target = os.path.abspath(os.path.join(base, rel_path))
-    if os.path.normcase(target).startswith(os.path.normcase(base + os.sep)):
-        return target
-    return None
+# ============================================================
+# 全局错误处理
+# ============================================================
 
 @app.errorhandler(413)
 def request_entity_too_large(error):
-    """文件大小超出Flask限制"""
+    """上传文件超出 MAX_CONTENT_LENGTH 限制时触发"""
     return jsonify({
         'success': False,
         'error': f'文件大小超出限制，最大允许 {settings.max_file_size_mb}MB'
@@ -59,7 +69,7 @@ def request_entity_too_large(error):
 
 @app.errorhandler(404)
 def not_found(error):
-    """页面未找到"""
+    """请求的路由不存在"""
     return jsonify({
         'success': False,
         'error': '请求的资源不存在'
@@ -68,714 +78,112 @@ def not_found(error):
 
 @app.errorhandler(500)
 def internal_error(error):
-    """服务器内部错误"""
+    """未捕获的服务端异常"""
     return jsonify({
         'success': False,
         'error': '服务器内部错误，请稍后重试'
     }), 500
 
 
-# 全局工作流图（带 MemorySaver，通过 thread_id 管理会话状态）
-workflow_graph = build_workflow()
-current_thread_id = None
-session_files = {}
-session_file_order = []
-cancelled_file_keys = set()
-session_lock = threading.Lock()
-
-
-def allowed_file(filename):
-    """检查文件扩展名是否允许"""
-    return PurePath(filename).suffix.lower().lstrip('.') in settings.allowed_extensions
-
-
-def _read_split_subject() -> Optional[str]:
-    """从 split_metadata.json 读取学科信息"""
-    meta_path = os.path.join(str(settings.results_dir), "split_metadata.json")
-    if os.path.exists(meta_path):
-        with open(meta_path, 'r', encoding='utf-8') as f:
-            return json.load(f).get("subject")
-    return None
-
-
-def _serialize_split_record(r) -> dict:
-    """将 SplitRecord ORM 对象序列化为前端 JSON 格式"""
-    return {
-        "id": r.id,
-        "subject": r.subject,
-        "model_provider": r.model_provider,
-        "file_names": json.loads(r.file_names_json) if r.file_names_json else [],
-        "question_count": r.question_count,
-        "created_at": r.created_at.isoformat() if r.created_at else None,
-    }
-
-
-def _serialize_question(q: Question) -> dict:
-    """将 Question ORM 对象序列化为前端 JSON 格式"""
-    subject = None
-    if q.batch:
-        subject = q.batch.subject
-    knowledge_tags = []
-    if q.tags:
-        for mapping in q.tags:
-            if mapping.tag:
-                knowledge_tags.append(mapping.tag.tag_name)
-
-    return {
-        'id': q.id,
-        'question_type': q.question_type,
-        'content_json': json.loads(q.content_json) if q.content_json else [],
-        'options_json': json.loads(q.options_json) if q.options_json else None,
-        'has_formula': q.has_formula,
-        'has_image': q.has_image,
-        'needs_correction': q.needs_correction,
-        'answer': q.answer,
-        'subject': subject,
-        'knowledge_tags': knowledge_tags,
-        'created_at': q.created_at.isoformat() if q.created_at else None,
-    }
-
-
-def _serialize_chat_session(session) -> dict:
-    """将 ChatSession ORM 对象序列化为前端 JSON"""
-    return {
-        'id': session.id,
-        'question_id': session.question_id,
-        'created_at': session.created_at.isoformat() if session.created_at else None,
-        'updated_at': session.updated_at.isoformat() if session.updated_at else None,
-    }
-
-
-def _serialize_question_detail(q: Question) -> dict:
-    """将 Question ORM 对象序列化为详情 JSON（含科目、标签、答案等）"""
-    base = _serialize_question(q)
-    # subject / knowledge_tags already set by _serialize_question
-    base['original_filename'] = q.batch.original_filename if q.batch else None
-    base['user_answer'] = q.user_answer
-    base['updated_at'] = q.updated_at.isoformat() if q.updated_at else None
-    base['review_status'] = q.review_status or '待复习'
-    base['image_refs_json'] = json.loads(q.image_refs_json) if q.image_refs_json else None
-    return base
-
-
 # ============================================================
-# 前端托管（生产模式：直接返回 Vite 构建产物）
-# ============================================================
-FRONTEND_DIST = os.path.join(settings.project_root, 'frontend', 'dist')
-
-
-
-
-# ============================================================
-# 认证工具
+# 全局中间件
 # ============================================================
 
 @app.before_request
 def require_login():
-    """所有 /api/ 路由（除 /api/auth/ 和 /api/status）要求登录"""
+    """登录鉴权：所有 /api/ 路由默认要求登录
+
+    豁免列表：
+      - /api/auth/*  — 登录、注册、获取当前用户等认证接口
+      - /api/status  — 系统状态查询（前端初始化时需要）
+    """
     if request.path.startswith('/api/'):
         if request.path.startswith('/api/auth/') or request.path == '/api/status':
             return None
         if 'user_id' not in session:
             return jsonify({'error': '请先登录', 'code': 'UNAUTHORIZED'}), 401
+        # 密码重置后失效旧 session：仅在写操作时查 DB 比对 session_version，
+        # GET 请求无需每次查库，降低高频只读接口的数据库压力
+        if request.method != 'GET':
+            user_id = session['user_id']
+            sess_ver = session.get('session_version')
+            with SessionLocal() as db:
+                user = crud.get_user_by_id(db, user_id)
+                if not user:
+                    session.clear()
+                    return jsonify({'error': '用户不存在', 'code': 'UNAUTHORIZED'}), 401
+                db_ver = user.session_version or 0
+                if sess_ver != db_ver:
+                    session.clear()
+                    return jsonify({'error': '登录已失效，请重新登录', 'code': 'SESSION_EXPIRED'}), 401
     return None
 
 
-@app.route('/api/auth/register', methods=['POST'])
-def auth_register():
-    """用户注册"""
-    data = request.get_json() or {}
-    email = (data.get('email') or '').strip().lower()
-    username = (data.get('username') or '').strip()
-    password = data.get('password') or ''
-
-    if not email or '@' not in email:
-        return jsonify({'error': '邮箱格式不正确'}), 400
-    if not username:
-        return jsonify({'error': '用户名不能为空'}), 400
-    if len(password) < 6:
-        return jsonify({'error': '密码至少 6 位'}), 400
-
-    with SessionLocal() as db:
-        if crud.get_user_by_email(db, email):
-            return jsonify({'error': '该邮箱已注册'}), 409
-        pwd_hash = generate_password_hash(password)
-        user = crud.create_user(db, email=email, password_hash=pwd_hash, username=username)
-        session['user_id'] = user.id
-        session['username'] = user.username
-        return jsonify({'user': {'id': user.id, 'email': user.email, 'username': user.username, 'is_admin': user.is_admin}}), 201
-
-
-@app.route('/api/auth/login', methods=['POST'])
-def auth_login():
-    """用户登录（支持邮箱或用户名）"""
-    data = request.get_json() or {}
-    identifier = (data.get('email') or data.get('identifier') or '').strip()
-    password = data.get('password') or ''
-
-    if not identifier:
-        return jsonify({'error': '请输入邮箱或用户名'}), 400
-
-    with SessionLocal() as db:
-        user = crud.get_user_by_login(db, identifier)
-        if not user or not check_password_hash(user.password_hash, password):
-            return jsonify({'error': '账号或密码错误'}), 401
-        session['user_id'] = user.id
-        session['username'] = user.username
-        return jsonify({'user': {'id': user.id, 'email': user.email, 'username': user.username, 'is_admin': user.is_admin}})
-
-
-@app.route('/api/auth/logout', methods=['POST'])
-def auth_logout():
-    """退出登录"""
-    session.clear()
-    return jsonify({'ok': True})
-
-
-@app.route('/api/auth/me', methods=['GET'])
-def auth_me():
-    """获取当前登录用户"""
-    user_id = session.get('user_id')
-    if not user_id:
-        return jsonify({'error': '未登录', 'code': 'UNAUTHORIZED'}), 401
-    with SessionLocal() as db:
-        user = crud.get_user_by_id(db, user_id)
-        if not user:
-            session.clear()
-            return jsonify({'error': '用户不存在', 'code': 'UNAUTHORIZED'}), 401
-        return jsonify({'user': {'id': user.id, 'email': user.email, 'username': user.username, 'is_admin': user.is_admin}})
-
-
-@app.route('/')
-def index():
-    """主页 - 返回 Vue SPA（落地页由前端路由 / 渲染）"""
-    return send_from_directory(FRONTEND_DIST, 'app.html')
-
-
-@app.route('/app.html')
-def app_page_redirect():
-    """旧路径重定向到规范 URL"""
-    return redirect('/app', code=301)
-
-
-@app.route('/app')
-@app.route('/app/<path:subpath>')
-def app_page(subpath=''):
-    """工作台及子路由 - 返回 Vue SPA"""
-    return send_from_directory(FRONTEND_DIST, 'app.html')
-
-
-@app.route('/auth')
-def auth_page():
-    """登录/注册页 - 返回 Vue SPA"""
-    return send_from_directory(FRONTEND_DIST, 'app.html')
-
-
-@app.route('/static/vue/<path:filename>')
-def serve_vue_dist(filename):
-    """提供 Vue 前端构建产物"""
-    return send_from_directory(FRONTEND_DIST, filename)
-
-
 # ============================================================
-# 遗留独立页面
+# 注册 Blueprint 路由
 # ============================================================
+# 各 Blueprint 定义在 routes/ 目录下：
+#   - auth.py      → /api/auth/*     用户认证
+#   - upload.py    → /api/*          文件上传、分割、导出
+#   - questions.py → /api/*          题目 CRUD、错题库、搜索
+#   - chat.py      → /api/*          AI 对话
+#   - stats.py     → /api/*          统计分析
+#   - settings.py  → /api/*          系统配置、模型管理
 
-
-@app.route('/record')
-def record_page():
-    """错题本记录页面"""
-    record_file = os.path.join(settings.project_root, 'record.html')
-    if os.path.exists(record_file):
-        return send_from_directory(settings.project_root, 'record.html')
-    return "记录页文件不存在", 404
+register_routes(app)
 
 
 # ============================================================
-# API 接口
+# 静态资源服务
 # ============================================================
+# 以下路由提供后端生成的文件资源，前端通过 Vite 代理访问：
+#   /download/*  — 导出的 Markdown 错题本文件
+#   /images/*    — PaddleOCR 解析出的图片（存储在 struct_dir/imgs/）
+#   /erased/*    — EnsExam 擦除手写字迹后的图片
 
+def _safe_join(base_dir: str, rel_path: str) -> str | None:
+    """安全路径拼接，防止目录遍历攻击（如 ../../etc/passwd）
 
-@app.route('/api/upload', methods=['POST'])
-def upload_file():
-    """处理文件上传（支持多文件）。
-
-    这里只负责把原始文件保存到 uploads 并登记到会话中；
-    标准化(PDF/图片 → 图片列表) + OCR + 分割，会在 /api/split 里由用户点击“开始分割”后统一触发。
-
-    Returns:
-        JSON响应，包含上传结果
-    """
-    # 支持多文件：前端用 'files' 字段发送，兼容单文件 'file' 字段
-    files = request.files.getlist('files')
-    if not files:
-        files = request.files.getlist('file')
-    if not files or all(f.filename == '' for f in files):
-        return jsonify({'error': '没有上传文件'}), 400
-
-    # 校验每个文件
-    for file in files:
-        if file.filename == '':
-            continue
-
-        if not allowed_file(file.filename):
-            return jsonify({
-                'error': f'不支持的文件格式: {file.filename}。支持: {", ".join(settings.allowed_extensions)}'
-            }), 400
-
-        file.seek(0, 2)
-        file_size = file.tell()
-        file.seek(0)
-        file_size_mb = file_size / (1024 * 1024)
-
-        if file_size_mb > settings.max_file_size_mb:
-            return jsonify({
-                'error': f'{file.filename} 大小为 {file_size_mb:.1f}MB，超出最大限制 {settings.max_file_size_mb}MB'
-            }), 400
-
-        if file_size == 0:
-            return jsonify({
-                'error': f'{file.filename} 为空文件，请重新选择'
-            }), 400
-
-    try:
-        global current_thread_id, session_files, session_file_order
-
-        file_keys = request.form.getlist('file_key')
-        if not file_keys:
-            file_keys = request.form.getlist('file_keys')
-
-        prepared = []
-        for i, file in enumerate(files):
-            if file.filename == '':
-                continue
-            fk = file_keys[i] if i < len(file_keys) and file_keys[i] else None
-            prepared.append((fk, file))
-
-        if not prepared:
-            return jsonify({'error': '没有上传文件'}), 400
-
-        results_out = []
-        for fk, file in prepared:
-            file_key = fk or f"{uuid.uuid4().hex}"
-
-            original_ext = os.path.splitext(file.filename)[1].lower().lstrip('.')
-            filename = f"{uuid.uuid4().hex}.{original_ext}"
-            filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
-            file.save(filepath)
-
-            with session_lock:
-                cancelled = file_key in cancelled_file_keys
-                if cancelled:
-                    cancelled_file_keys.discard(file_key)
-
-            if cancelled:
-                try:
-                    os.remove(filepath)
-                except FileNotFoundError:
-                    pass
-                continue
-
-            with session_lock:
-                if current_thread_id is not None:
-                    current_thread_id = None
-                    session_files = {}
-                    session_file_order = []
-
-                session_files[file_key] = {
-                    "filename": file.filename,
-                    "filepath": filepath,
-                }
-                if file_key not in session_file_order:
-                    session_file_order.append(file_key)
-
-            results_out.append({
-                "file_key": file_key,
-                "filename": file.filename,
-            })
-
-        return jsonify({
-            'success': True,
-            'message': '上传成功',
-            'result': {
-                'file_count': len(results_out),
-                'files': results_out,
-            }
-        })
-
-    except FileNotFoundError:
-        return jsonify({
-            'success': False,
-            'error': '文件保存失败，请重新上传'
-        }), 500
-    except Exception as e:
-        logger.exception("文件处理失败")
-        return jsonify({
-            'success': False,
-            'error': '文件处理失败，请稍后重试'
-        }), 500
-
-
-@app.route('/api/cancel_file', methods=['POST'])
-def cancel_file():
-    try:
-        global current_thread_id, session_files, session_file_order
-
-        data = request.get_json(silent=True) or {}
-        file_key = data.get('file_key')
-        if not file_key:
-            return jsonify({
-                'success': False,
-                'error': '缺少 file_key'
-            }), 400
-
-        if current_thread_id is not None:
-            return jsonify({
-                'success': False,
-                'error': '已开始分割，无法撤销单个文件；请重置后重新上传'
-            }), 400
-
-        filepath = None
-        existed = False
-        with session_lock:
-            cancelled_file_keys.add(file_key)
-
-            existed = file_key in session_files
-            v = session_files.pop(file_key, None) or {}
-            filepath = v.get('filepath')
-            if file_key in session_file_order:
-                session_file_order.remove(file_key)
-
-            cancelled_file_keys.discard(file_key)
-
-        if filepath:
-            try:
-                os.remove(filepath)
-            except FileNotFoundError:
-                pass
-
-        return jsonify({
-            'success': True,
-            'message': '已撤销该文件' if existed else '已标记撤销该文件',
-        })
-
-    except Exception as e:
-        logger.exception("撤销文件失败")
-        return jsonify({
-            'success': False,
-            'error': '撤销失败，请稍后重试'
-        }), 500
-
-
-@app.route('/api/split', methods=['POST'])
-def split_questions():
-    """开始执行标准化 + OCR + 分割。
-
-    用户点击前端“开始分割题目”后调用该接口：
-    - 标准化输入（PDF/图片 → 图片列表）
-    - 触发 Agent/OCR 并分割题目
-
-    Returns:
-        JSON响应，包含分割后的题目
-    """
-    try:
-        global workflow_graph, current_thread_id, session_files, session_file_order
-
-        # 读取请求体参数（模型供应商 + 模型名称）
-        data = request.get_json(silent=True) or {}
-        model_provider = data.get("model_provider", "openai")
-        model_name = data.get("model_name")  # 可选，None 时使用 provider 默认模型
-        if not settings.is_valid_provider(model_provider):
-            return jsonify({
-                'success': False,
-                'error': f'不支持的模型供应商: {model_provider}'
-            }), 400
-
-        # 从数据库加载用户的 LLM + OCR 凭据
-        user_id = session.get('user_id')
-        ocr_credentials = {}
-        if user_id:
-            settings.load_providers_from_db(user_id)
-            with SessionLocal() as db:
-                ocr_provider = crud.get_active_provider(db, user_id, 'paddleocr')
-                if ocr_provider:
-                    ocr_credentials = {
-                        "api_url": ocr_provider.base_url,
-                        "token": ocr_provider.api_key,
-                        "model": ocr_provider.model_name,
-                        "use_doc_orientation": ocr_provider.use_doc_orientation,
-                        "use_doc_unwarping": ocr_provider.use_doc_unwarping,
-                        "use_chart_recognition": ocr_provider.use_chart_recognition,
-                    }
-
-        with session_lock:
-            keys = list(session_file_order)
-            file_paths = []
-            for k in keys:
-                v = session_files.get(k) or {}
-                fp = v.get('filepath')
-                if fp:
-                    file_paths.append(fp)
-
-        if not file_paths:
-            return jsonify({
-                'success': False,
-                'error': '请先上传文件'
-            }), 400
-
-        # 擦除手写字迹（EnsExam 已接入且前端未关闭擦除开关时执行）
-        erase = data.get("erase", True)
-        ensexam_configured = settings.model_path.exists()
-        if ensexam_configured and erase:
-            try:
-                from models.inference import InferenceEngine
-                engine = InferenceEngine()
-                erased_paths = []
-                for fp in file_paths:
-                    # 如果是 PDF 转换为图片的临时路径，或者是直接上传的图片
-                    with open(fp, 'rb') as f:
-                        img_bytes = f.read()
-                    result_img = engine.run(img_bytes)
-                    
-                    # 生成擦除后的文件名，保存在 erased_dir
-                    erased_name = f"auto_{uuid.uuid4().hex[:8]}_{os.path.basename(fp)}"
-                    if not erased_name.lower().endswith(('.png', '.jpg', '.jpeg')):
-                        erased_name += '.png'
-                    
-                    erased_path = settings.erased_dir / erased_name
-                    result_img.save(str(erased_path), format='PNG')
-                    erased_paths.append(str(erased_path))
-                
-                file_paths = erased_paths
-                logger.info("EnsExam 已接入，自动擦除 %d 张图片的手写字迹", len(erased_paths))
-            except Exception:
-                logger.exception("自动擦除手写字迹失败，使用原图继续流程")
-
-        current_thread_id = str(uuid.uuid4())
-        config = {"configurable": {"thread_id": current_thread_id}}
-
-        initial_state = {
-            "file_paths": file_paths,
-            "model_provider": model_provider,
-            "model_name": model_name,
-            "ocr_credentials": ocr_credentials,
-        }
-        workflow_graph.invoke(initial_state, config=config)
-        state = workflow_graph.invoke(None, config=config)
-
-        questions = state.get('questions', [])
-        warnings = state.get('warnings', [])
-
-        # 自动保存分割记录
-        try:
-            subject = _read_split_subject()
-
-            with session_lock:
-                file_names = [
-                    session_files.get(k, {}).get("filename", "未知")
-                    for k in session_file_order
-                ]
-
-            with SessionLocal() as db:
-                crud.save_split_record(db, subject, model_provider, file_names, questions, user_id=session.get('user_id'))
-        except Exception:
-            logger.warning("保存分割记录失败，不影响主流程", exc_info=True)
-
-        return jsonify({
-            'success': True,
-            'message': f'成功分割 {len(questions)} 道题目',
-            'questions': questions,
-            'warnings': warnings,
-        })
-
-    except Exception as e:
-        logger.exception("题目分割失败")
-        return jsonify({
-            'success': False,
-            'error': '题目分割失败，请稍后重试'
-        }), 500
-
-
-@app.route('/api/split-records', methods=['GET'])
-def get_split_records():
-    """获取最近 N 次分割历史记录，limit 由前端通过查询参数指定"""
-    try:
-        limit = request.args.get('limit', 10, type=int)
-        limit = max(1, min(limit, crud.MAX_SPLIT_RECORDS))  # 上限与保留条数一致
-
-        with SessionLocal() as db:
-            records = crud.get_recent_split_records(db, limit, user_id=session.get('user_id'))
-            result = [_serialize_split_record(r) for r in records]
-
-        return jsonify({"success": True, "records": result})
-
-    except Exception as e:
-        logger.exception("获取分割记录失败")
-        return jsonify({"success": False, "error": "获取分割记录失败"}), 500
-
-
-@app.route('/api/split-records/<int:record_id>', methods=['GET'])
-def get_split_record_detail(record_id):
-    """获取单条分割记录的完整数据（含 questions）"""
-    try:
-        with SessionLocal() as db:
-            record = crud.get_split_record_by_id(db, record_id)
-            if not record:
-                return jsonify({"success": False, "error": "记录不存在"}), 404
-
-            result = _serialize_split_record(record)
-            result["questions"] = json.loads(record.questions_json) if record.questions_json else []
-
-        return jsonify({"success": True, "record": result})
-
-    except Exception as e:
-        logger.exception("获取分割记录详情失败")
-        return jsonify({"success": False, "error": "获取分割记录详情失败"}), 500
-
-
-@app.route('/api/export', methods=['POST'])
-def export_wrongbook():
-    """
-    注入选中题目 ID 并恢复图执行导出
-
-    图执行: export → END
-
-    Returns:
-        JSON响应，包含导出文件路径
-    """
-    try:
-        global current_thread_id, session_files, session_file_order
-
-        data = request.get_json(silent=True) or {}
-        selected_ids = data.get('selected_ids', [])
-
-        if not isinstance(selected_ids, list):
-            return jsonify({
-                'success': False,
-                'error': 'selected_ids 必须为列表'
-            }), 400
-
-        if not selected_ids:
-            return jsonify({
-                'success': False,
-                'error': '未选择任何题目'
-            }), 400
-
-        # 检查是否已分割（通过 questions.json 存在性判断，不依赖内存中的 current_thread_id）
-        results_dir = settings.results_dir
-        questions_file = os.path.join(results_dir, "questions.json")
-        if not os.path.exists(questions_file):
-            return jsonify({
-                'success': False,
-                'error': '请先分割题目'
-            }), 400
-
-        with open(questions_file, 'r', encoding='utf-8') as f:
-            questions = json.load(f)
-
-        output_path = export_wrongbook_md(
-            questions,
-            selected_ids,
-        )
-
-        return jsonify({
-            'success': True,
-            'message': '错题本导出成功',
-            'output_path': output_path
-        })
-
-    except Exception as e:
-        logger.exception("导出失败")
-        return jsonify({
-            'success': False,
-            'error': '导出失败，请稍后重试'
-        }), 500
-
-
-@app.route('/api/questions', methods=['GET'])
-def get_questions():
-    """
-    获取已分割的题目列表
-
-    Returns:
-        JSON响应，包含题目列表
+    将 rel_path 拼接到 base_dir 后，检查结果路径是否仍在 base_dir 内。
+    如果路径跳出了 base_dir，返回 None。
     """
-    try:
-        results_dir = settings.results_dir
-        questions_file = os.path.join(results_dir, "questions.json")
-
-        if not os.path.exists(questions_file):
-            return jsonify({
-                'success': True,
-                'questions': [],
-                'message': '暂无题目数据'
-            })
-
-        with open(questions_file, 'r', encoding='utf-8') as f:
-            questions = json.load(f)
-
-        return jsonify({
-            'success': True,
-            'questions': questions
-        })
-
-    except json.JSONDecodeError:
-        return jsonify({
-            'success': False,
-            'error': '题目数据文件格式错误，请重新分割题目'
-        }), 500
-    except Exception as e:
-        logger.exception("获取题目列表失败")
-        return jsonify({
-            'success': False,
-            'error': '获取题目列表失败，请稍后重试'
-        }), 500
-
-
-@app.route('/preview')
-def preview():
-    """显示预览页面"""
-    results_dir = settings.results_dir
-    preview_file = os.path.join(results_dir, "preview.html")
-
-    if os.path.exists(preview_file):
-        with open(preview_file, 'r', encoding='utf-8') as f:
-            content = f.read()
-        return content
-    else:
-        return "预览文件不存在，请先分割题目", 404
+    base = os.path.abspath(base_dir)
+    target = os.path.abspath(os.path.join(base, rel_path))
+    if os.path.normcase(target).startswith(os.path.normcase(base + os.sep)):
+        return target
+    return None
 
 
-# ============================================================
-# 资源服务
-# ============================================================
+@app.route('/uploads/<path:filename>')
+def serve_upload(filename):
+    """提供用户上传的原始文件（笔记图片等）"""
+    file_path = _safe_join(str(settings.upload_dir), filename)
+    if not file_path or not os.path.exists(file_path):
+        return jsonify({'success': False, 'error': '文件不存在'}), 404
+    return send_file(file_path)
 
 
 @app.route('/download/<path:filename>')
 def download_file(filename):
-    """下载结果文件"""
-    results_dir = settings.results_dir
-    file_path = _safe_join(results_dir, filename)
-    if not file_path:
-        return jsonify({
-            'success': False,
-            'error': '非法文件路径'
-        }), 400
+    """下载结果文件（如导出的 Markdown 错题本）
 
+    禁用浏览器缓存，确保每次下载都是最新版本。
+    """
+    file_path = _safe_join(settings.results_dir, filename)
+    if not file_path:
+        return jsonify({'success': False, 'error': '非法文件路径'}), 400
     if not os.path.exists(file_path):
-        return jsonify({
-            'success': False,
-            'error': '文件不存在'
-        }), 404
+        return jsonify({'success': False, 'error': '文件不存在'}), 404
 
     resp = send_file(
         file_path,
-        as_attachment=True,
-        download_name=os.path.basename(filename),
+        as_attachment=True,                          # 触发浏览器下载而非预览
+        download_name=os.path.basename(filename),    # 下载时显示的文件名
         conditional=False,
         etag=False,
         max_age=0,
     )
+    # 强制禁用缓存，避免浏览器返回旧文件
     resp.headers['Cache-Control'] = 'no-store, no-cache, must-revalidate, max-age=0'
     resp.headers['Pragma'] = 'no-cache'
     resp.headers['Expires'] = '0'
@@ -784,7 +192,11 @@ def download_file(filename):
 
 @app.route('/images/<path:filename>')
 def serve_image(filename):
-    """提供 OCR 解析出的图片资源"""
+    """提供 PaddleOCR 解析出的图片资源
+
+    OCR 处理后的结构化图片存储在 struct_dir/imgs/ 目录下，
+    前端题目卡片中的图片通过此接口加载。
+    """
     base = os.path.join(settings.struct_dir, "imgs")
     file_path = _safe_join(base, filename)
     if not file_path or not os.path.exists(file_path):
@@ -794,1108 +206,39 @@ def serve_image(filename):
 
 @app.route('/erased/<path:filename>')
 def serve_erased_image(filename):
-    """提供擦除结果图片"""
+    """提供 EnsExam 擦除手写字迹后的图片
+
+    用户上传试卷后，EnsExam 模型会擦除手写笔迹，
+    还原干净的题目底图，结果保存在 erased_dir。
+    """
     file_path = _safe_join(str(settings.erased_dir), filename)
     if not file_path or not os.path.exists(file_path):
         return jsonify({'success': False, 'error': '图片不存在'}), 404
     return send_file(file_path)
 
 
-@app.route('/api/models/erase', methods=['POST'])
-def erase_text():
-    """文字擦除接口
-
-    使用 GAN 模型将试卷图片中的手写笔迹擦除，还原干净的题目底图。
-
-    Request:
-        multipart/form-data，字段 file：图片文件（JPEG/PNG/BMP 等）
-
-    Response:
-        {"success": true, "result_url": "/erased/<filename>"}
-    """
-    if 'file' not in request.files:
-        return jsonify({'success': False, 'error': '缺少 file 字段'}), 400
-
-    file = request.files['file']
-    if not file or not file.filename:
-        return jsonify({'success': False, 'error': '文件为空'}), 400
-
-    ext = os.path.splitext(file.filename)[1].lower() or '.png'
-    allowed = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.webp'}
-    if ext not in allowed:
-        return jsonify({'success': False, 'error': f'不支持的图片格式: {ext}'}), 400
-
-    try:
-        image_bytes = file.read()
-    except Exception:
-        return jsonify({'success': False, 'error': '读取文件失败'}), 400
-
-    try:
-        from models.inference import InferenceEngine
-
-        result_img = InferenceEngine().run(image_bytes)
-
-        filename = uuid.uuid4().hex + '.png'
-        save_path = settings.erased_dir / filename
-        result_img.save(save_path, format='PNG')
-
-        return jsonify({'success': True, 'result_url': f'/erased/{filename}'})
-
-    except FileNotFoundError as e:
-        logger.error("模型权重缺失: %s", e)
-        return jsonify({'success': False, 'error': str(e)}), 503
-    except Exception:
-        logger.exception("文字擦除失败")
-        return jsonify({'success': False, 'error': '擦除处理失败，请稍后重试'}), 500
-
-
-@app.route('/api/status', methods=['GET'])
-def get_status():
-    """
-    获取系统状态
-
-    Returns:
-        JSON响应，包含系统配置和状态
-    """
-    try:
-        user_id = session.get('user_id')
-
-        # 检查 EnsExam 模型权重是否存在
-        ensexam_configured = settings.model_path.exists()
-
-        if user_id:
-            # 已登录：从数据库读取用户配置
-            with SessionLocal() as db:
-                paddle_provider = crud.get_active_provider(db, user_id, 'paddleocr')
-                paddleocr_configured = bool(paddle_provider and paddle_provider.api_key and paddle_provider.base_url)
-
-                # 构建用户级可用模型列表
-                available_models = []
-                for category, label in [('openai', 'OpenAI'), ('anthropic', 'Anthropic')]:
-                    provider = crud.get_active_provider(db, user_id, category)
-                    if provider and provider.api_key:
-                        available_models.append({
-                            'value': category,
-                            'label': provider.name or label,
-                            'configured': True,
-                            'status': '配置成功',
-                            'default_model': provider.model_name or '',
-                            'models': [provider.model_name] if provider.model_name else [],
-                        })
-                    else:
-                        available_models.append({
-                            'value': category,
-                            'label': label,
-                            'configured': False,
-                            'status': '未配置',
-                            'default_model': '',
-                            'models': [],
-                        })
-        else:
-            # 未登录：返回空状态
-            paddleocr_configured = False
-            available_models = []
-
-        status = {
-            'paddleocr_configured': paddleocr_configured,
-            'ensexam_configured': ensexam_configured,
-            'available_models': available_models,
-            'langsmith_enabled': os.getenv('LANGSMITH_TRACING', 'false').lower() == 'true',
-            'output_dirs': {
-                'pages': str(settings.pages_dir),
-                'struct': str(settings.struct_dir),
-                'results': str(settings.results_dir),
-            }
-        }
-
-        return jsonify({
-            'success': True,
-            'status': status
-        })
-
-    except Exception as e:
-        logger.exception("获取系统状态失败")
-        return jsonify({
-            'success': False,
-            'error': '获取系统状态失败，请稍后重试'
-        }), 500
-
-
-@app.route('/api/config', methods=['GET'])
-def get_config():
-    """获取当前用户的 provider 配置"""
-    try:
-        user_id = session.get('user_id')
-        if not user_id:
-            return jsonify({'success': False, 'error': '请先登录'}), 401
-        with SessionLocal() as db:
-            config = crud.get_user_providers(db, user_id)
-        return jsonify({'success': True, 'config': config})
-    except Exception as e:
-        logger.exception("获取配置失败")
-        return jsonify({'success': False, 'error': '获取配置失败'}), 500
-
-
-@app.route('/api/config', methods=['PUT'])
-def update_config():
-    """保存当前用户的 provider 配置到数据库"""
-    try:
-        data = request.get_json(force=True) or {}
-        user_id = session.get('user_id')
-
-        if not user_id:
-            return jsonify({'success': False, 'error': '请先登录'}), 401
-
-        with SessionLocal() as db:
-            try:
-                crud.save_user_providers(db, user_id, data)
-            except Exception:
-                db.rollback()
-                raise
-        return jsonify({'success': True})
-
-    except Exception as e:
-        logger.exception("更新配置失败")
-        return jsonify({'success': False, 'error': '更新配置失败，请检查日志'}), 500
-
-
-@app.route('/api/models/list', methods=['POST'])
-def list_models():
-    """代理请求目标 API 的模型列表，绕过浏览器 CORS 限制。
-
-    请求体:
-        type: 'openai' | 'anthropic'
-        api_key: API Key（可选，留空则使用系统已配置的 key）
-        base_url: Base URL（可选）
-        provider_id: 已有 provider 的 id（可选，用于回退读取已存 key）
-    """
-    try:
-        data = request.get_json(force=True) or {}
-        provider_type = data.get('type', 'openai')
-        api_key = data.get('api_key') or ''
-        base_url = data.get('base_url') or ''
-
-        # 如果前端没传 key，从数据库读取已保存的配置
-        if not api_key:
-            user_id = session.get('user_id')
-            provider_id = data.get('provider_id')
-            if user_id:
-                with SessionLocal() as db:
-                    if provider_id:
-                        provider = db.query(crud.ProviderConfig).filter_by(
-                            id=provider_id, user_id=user_id
-                        ).first()
-                    else:
-                        provider = crud.get_active_provider(db, user_id, provider_type)
-                    if provider:
-                        if not api_key:
-                            api_key = provider.api_key or ''
-                        if not base_url:
-                            base_url = provider.base_url or ''
-
-        if not api_key:
-            return jsonify({'error': '未提供 API Key，请先在设置中配置'}), 400
-
-        if provider_type == 'openai':
-            url = (base_url.rstrip('/') if base_url else 'https://api.openai.com') + '/v1/models'
-            headers = {'Authorization': f'Bearer {api_key}'}
-            resp = http_requests.get(url, headers=headers, timeout=15)
-
-            # 部分 OpenAI 兼容 API 的路径可能不带 /v1
-            if resp.status_code == 404:
-                url_alt = (base_url.rstrip('/') if base_url else 'https://api.openai.com') + '/models'
-                resp = http_requests.get(url_alt, headers=headers, timeout=15)
-
-            if resp.status_code != 200:
-                return jsonify({'error': f'API 返回 {resp.status_code}: {resp.text[:200]}'}), 502
-
-            body = resp.json()
-            models = sorted([m['id'] for m in body.get('data', [])]) if 'data' in body else []
-            return jsonify({'models': models})
-
-        elif provider_type == 'anthropic':
-            # Anthropic 没有官方 list models 接口，返回常用模型列表
-            models = [
-                'claude-opus-4-20250514',
-                'claude-sonnet-4-20250514',
-                'claude-haiku-4-20250506',
-                'claude-3-5-sonnet-20241022',
-                'claude-3-5-haiku-20241022',
-                'claude-3-opus-20240229',
-            ]
-            return jsonify({'models': models})
-
-        else:
-            return jsonify({'error': f'不支持的 provider 类型: {provider_type}'}), 400
-
-    except http_requests.Timeout:
-        return jsonify({'error': '请求超时，请检查 Base URL 是否正确'}), 504
-    except http_requests.ConnectionError:
-        return jsonify({'error': '无法连接目标 API，请检查 Base URL'}), 502
-    except Exception as e:
-        logger.exception("获取模型列表失败")
-        return jsonify({'error': str(e)}), 500
-
-
-@app.route('/api/paddleocr/test', methods=['POST'])
-def test_paddleocr():
-    """测试 PaddleOCR API Token 是否可用。
-
-    请求体:
-        api_token: API Token（可选，留空则从数据库读取）
-        api_url: API URL（可选，留空则从数据库读取）
-        provider_id: 已有 provider 的 id（可选，用于回退读取已存 token）
-    """
-    try:
-        data = request.get_json(force=True) or {}
-        api_token = data.get('api_token') or ''
-        api_url = data.get('api_url') or ''
-
-        # 前端未提供凭据时，从数据库读取已保存的配置
-        if not api_token or not api_url:
-            user_id = session.get('user_id')
-            provider_id = data.get('provider_id')
-            if user_id:
-                with SessionLocal() as db:
-                    if provider_id:
-                        provider = db.query(crud.ProviderConfig).filter_by(
-                            id=provider_id, user_id=user_id
-                        ).first()
-                    else:
-                        provider = crud.get_active_provider(db, user_id, 'paddleocr')
-                    if provider:
-                        if not api_token:
-                            api_token = provider.api_key or ''
-                        if not api_url:
-                            api_url = provider.base_url or ''
-
-        if not api_token or not api_url:
-            return jsonify({'error': '未提供 API Token 或 API URL'}), 400
-
-        # 用 GET 请求 API URL，带 token 验证认证是否有效
-        headers = {'Authorization': f'bearer {api_token}'}
-        resp = http_requests.get(api_url, headers=headers, timeout=10)
-
-        # 401/403 表示 token 无效
-        if resp.status_code in (401, 403):
-            return jsonify({'success': False, 'error': '认证失败，请检查 API Token 是否正确'})
-
-        # 其他非 5xx 状态码都认为 token 有效（包括 404/405 等，因为 GET 可能不被支持但认证通过了）
-        if resp.status_code >= 500:
-            return jsonify({'success': False, 'error': f'服务端错误 ({resp.status_code})，请检查 API URL'})
-
-        return jsonify({'success': True, 'message': 'API Token 验证通过'})
-
-    except http_requests.Timeout:
-        return jsonify({'success': False, 'error': '请求超时，请检查 API URL 是否正确'}), 504
-    except http_requests.ConnectionError:
-        return jsonify({'success': False, 'error': '无法连接，请检查 API URL'}), 502
-    except Exception as e:
-        logger.exception("测试 PaddleOCR 连接失败")
-        return jsonify({'success': False, 'error': str(e)}), 500
-
-
-@app.route('/api/history', methods=['GET'])
-def get_history():
-    """
-    分页查询历史题目（全部题目，非仅错题）
-
-    Query参数:
-        page: 页码（默认1）
-        page_size: 每页数量（默认20）
-        start_date: 开始日期（可选，格式：YYYY-MM-DD）
-        end_date: 结束日期（可选，格式：YYYY-MM-DD）
-    """
-    try:
-        page = max(1, request.args.get('page', 1, type=int))
-        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
-        start_date_str = request.args.get('start_date')
-        end_date_str = request.args.get('end_date')
-
-        # 解析日期
-        start_date = None
-        end_date = None
-        if start_date_str:
-            start_date = datetime.strptime(start_date_str, '%Y-%m-%d')
-        if end_date_str:
-            end_date = datetime.strptime(end_date_str, '%Y-%m-%d')
-
-        with SessionLocal() as db:
-            questions, total = crud.get_history_questions(
-                db,
-                start_date=start_date,
-                end_date=end_date,
-                page=page,
-                page_size=page_size
-            )
-
-            total_pages = (total + page_size - 1) // page_size
-            items = [_serialize_question(q) for q in questions]
-
-            return jsonify({
-                'success': True,
-                'items': items,
-                'total': total,
-                'page': page,
-                'page_size': page_size,
-                'total_pages': total_pages
-            })
-
-    except ValueError:
-        return jsonify({'success': False, 'error': '日期格式错误，请使用 YYYY-MM-DD 格式'}), 400
-    except Exception as e:
-        logger.exception("查询历史题目失败")
-        return jsonify({'success': False, 'error': '查询失败，请稍后重试'}), 500
-
-
-@app.route('/api/search', methods=['GET'])
-def search_questions():
-    """
-    搜索题目（知识点/题型/关键字）
-
-    Query参数:
-        keyword: 关键字搜索（匹配题目内容）
-        knowledge_tag: 知识点标签筛选
-        question_type: 题型筛选
-        page: 页码（默认1）
-        page_size: 每页数量（默认20）
-    """
-    try:
-        page = max(1, request.args.get('page', 1, type=int))
-        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
-        keyword = request.args.get('keyword', type=str)
-        knowledge_tag = request.args.get('knowledge_tag', type=str)
-        question_type = request.args.get('question_type', type=str)
-
-        # 至少需要一个搜索条件
-        if not any([keyword, knowledge_tag, question_type]):
-            return jsonify({
-                'success': False,
-                'error': '至少需要提供一个搜索条件（keyword/knowledge_tag/question_type）'
-            }), 400
-
-        with SessionLocal() as db:
-            questions, total = crud.search_questions(
-                db,
-                keyword=keyword if keyword else None,
-                knowledge_tag=knowledge_tag if knowledge_tag else None,
-                question_type=question_type if question_type else None,
-                page=page,
-                page_size=page_size
-            )
-
-            total_pages = (total + page_size - 1) // page_size
-            items = [_serialize_question(q) for q in questions]
-
-            return jsonify({
-                'success': True,
-                'items': items,
-                'total': total,
-                'page': page,
-                'page_size': page_size,
-                'total_pages': total_pages
-            })
-
-    except Exception as e:
-        logger.exception("搜索题目失败")
-        return jsonify({'success': False, 'error': '搜索失败，请稍后重试'}), 500
-
-
-@app.route('/api/stats', methods=['GET'])
-def get_stats():
-    """
-    获取知识点统计信息
-    """
-    try:
-        subject = request.args.get('subject', type=str) or None
-        with SessionLocal() as db:
-            stats = crud.get_knowledge_stats(db, subject=subject, user_id=session.get('user_id'))
-            return jsonify({
-                'success': True,
-                'stats': stats,
-                'total_tags': len(stats)
-            })
-
-    except Exception as e:
-        logger.exception("获取统计失败")
-        return jsonify({'success': False, 'error': '获取统计失败，请稍后重试'}), 500
-
-
-@app.route('/api/question/<int:question_id>', methods=['DELETE'])
-def delete_question(question_id):
-    """
-    删除题目
-
-    Path参数:
-        question_id: 题目ID
-    """
-    try:
-        with SessionLocal() as db:
-            success = crud.delete_question(db, question_id)
-            if success:
-                return jsonify({
-                    'success': True,
-                    'message': '题目已删除'
-                })
-            else:
-                return jsonify({
-                    'success': False,
-                    'error': '题目不存在'
-                }), 404
-
-    except Exception as e:
-        logger.exception("删除题目失败")
-        return jsonify({'success': False, 'error': '删除失败，请稍后重试'}), 500
-
-
 # ============================================================
-# 错题库 API
+# 启动入口
 # ============================================================
 
-
-@app.route('/api/error-bank', methods=['GET'])
-def get_error_bank():
-    """
-    错题库统一查询
-
-    Query参数:
-        subject: 科目筛选
-        knowledge_tag: 知识点标签筛选
-        question_type: 题型筛选
-        keyword: 关键字搜索
-        start_date: 开始日期（YYYY-MM-DD）
-        end_date: 结束日期（YYYY-MM-DD）
-        page: 页码（默认1）
-        page_size: 每页数量（默认20）
-    """
-    try:
-        page = max(1, request.args.get('page', 1, type=int))
-        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
-        subject = request.args.get('subject', type=str) or None
-        knowledge_tag = request.args.get('knowledge_tag', type=str) or None
-        question_type = request.args.get('question_type', type=str) or None
-        keyword = request.args.get('keyword', type=str) or None
-        review_status = request.args.get('review_status', type=str) or None
-        start_date_str = request.args.get('start_date')
-        end_date_str = request.args.get('end_date')
-
-        start_date = None
-        end_date = None
-        if start_date_str:
-            start_date = datetime.strptime(start_date_str, '%Y-%m-%d')
-        if end_date_str:
-            end_date = datetime.strptime(end_date_str, '%Y-%m-%d')
-
-        with SessionLocal() as db:
-            questions, total = crud.query_questions(
-            db,
-            user_id=session.get('user_id'),
-                subject=subject,
-                knowledge_tag=knowledge_tag,
-                question_type=question_type,
-                keyword=keyword,
-                start_date=start_date,
-                end_date=end_date,
-                review_status=review_status,
-                page=page,
-                page_size=page_size,
-            )
-
-            total_pages = (total + page_size - 1) // page_size
-            items = [_serialize_question_detail(q) for q in questions]
-
-            return jsonify({
-                'success': True,
-                'items': items,
-                'total': total,
-                'page': page,
-                'page_size': page_size,
-                'total_pages': total_pages,
-            })
-
-    except ValueError:
-        return jsonify({'success': False, 'error': '日期格式错误，请使用 YYYY-MM-DD 格式'}), 400
-    except Exception as e:
-        logger.exception("查询错题库失败")
-        return jsonify({'success': False, 'error': '查询失败，请稍后重试'}), 500
-
-
-@app.route('/api/subjects', methods=['GET'])
-def get_subjects():
-    """获取所有科目列表"""
-    try:
-        with SessionLocal() as db:
-            subjects = crud.get_existing_subjects(db, user_id=session.get('user_id'))
-            return jsonify({'success': True, 'subjects': subjects})
-    except Exception as e:
-        logger.exception("获取科目列表失败")
-        return jsonify({'success': False, 'error': '获取科目列表失败'}), 500
-
-
-@app.route('/api/question-types', methods=['GET'])
-def get_question_types():
-    """获取所有题型列表"""
-    try:
-        with SessionLocal() as db:
-            types = crud.get_existing_question_types(db, user_id=session.get('user_id'))
-            return jsonify({'success': True, 'question_types': types})
-    except Exception as e:
-        logger.exception("获取题型列表失败")
-        return jsonify({'success': False, 'error': '获取题型列表失败'}), 500
-
-
-@app.route('/api/question/<int:question_id>', methods=['PATCH'])
-def update_question(question_id):
-    """编辑题目内容和/或答案"""
-    try:
-        data = request.get_json(silent=True) or {}
-        with SessionLocal() as db:
-            question = db.get(Question, question_id)
-            if not question:
-                return jsonify({'success': False, 'error': '题目不存在'}), 404
-            try:
-                if 'content' in data:
-                    text = str(data['content'])[:20000]
-                    blocks = [{'block_type': 'text', 'content': text}]
-                    question.content_json = json.dumps(blocks, ensure_ascii=False)
-                    question.content_hash = crud.compute_content_hash(blocks)
-                if 'answer' in data:
-                    question.answer = str(data['answer'])[:10000] if data['answer'] else None
-                question.updated_at = datetime.utcnow()
-                db.commit()
-                return jsonify({'success': True})
-            except Exception:
-                db.rollback()
-                raise
-    except Exception:
-        logger.exception("编辑题目失败")
-        return jsonify({'success': False, 'error': '保存失败，请稍后重试'}), 500
-
-
-@app.route('/api/question/<int:question_id>/answer', methods=['PATCH'])
-def update_question_answer(question_id):
-    """保存用户答案"""
-    try:
-        data = request.get_json(silent=True) or {}
-        user_answer = data.get('user_answer')
-        if user_answer is None:
-            return jsonify({'success': False, 'error': '缺少 user_answer 字段'}), 400
-        if len(user_answer) > 10000:
-            return jsonify({'success': False, 'error': '答案内容过长（最多 10000 字符）'}), 400
-
-        with SessionLocal() as db:
-            question = crud.update_user_answer(db, question_id, user_answer)
-            if not question:
-                return jsonify({'success': False, 'error': '题目不存在'}), 404
-
-            return jsonify({
-                'success': True,
-                'message': '答案已保存',
-                'user_answer': question.user_answer,
-                'updated_at': question.updated_at.isoformat() if question.updated_at else None,
-            })
-
-    except Exception as e:
-        logger.exception("保存答案失败")
-        return jsonify({'success': False, 'error': '保存答案失败，请稍后重试'}), 500
-
-
-@app.route('/api/question/<int:question_id>/review-status', methods=['PATCH'])
-def update_question_review_status(question_id):
-    """更新题目复习状态"""
-    try:
-        data = request.get_json(silent=True) or {}
-        review_status = data.get('review_status')
-        if not review_status:
-            return jsonify({'success': False, 'error': '缺少 review_status 字段'}), 400
-
-        with SessionLocal() as db:
-            question = crud.update_review_status(db, question_id, review_status)
-            if not question:
-                return jsonify({'success': False, 'error': '题目不存在'}), 404
-
-            return jsonify({
-                'success': True,
-                'message': '复习状态已更新',
-                'review_status': question.review_status,
-                'updated_at': question.updated_at.isoformat() if question.updated_at else None,
-            })
-
-    except ValueError:
-        return jsonify({'success': False, 'error': f'无效的复习状态，可选值: {", ".join(crud.VALID_REVIEW_STATUSES)}'}), 400
-    except Exception as e:
-        logger.exception("更新复习状态失败")
-        return jsonify({'success': False, 'error': '更新复习状态失败，请稍后重试'}), 500
-
-
-@app.route('/api/dashboard-stats', methods=['GET'])
-def get_dashboard_stats():
-    """获取 Dashboard 所需的完整统计数据，支持 ?subject= 学科筛选"""
-    try:
-        subject = request.args.get('subject') or None
-        uid = session.get('user_id')
-        with SessionLocal() as db:
-            # 可用学科列表（按当前用户隔离）
-            subjects = crud.get_existing_subjects(db, user_id=uid)
-
-            # 复习状态统计
-            review_stats = crud.get_review_status_stats(db, subject=subject, user_id=uid)
-
-            # 总体统计
-            statistics = crud.get_statistics(db, subject=subject, user_id=uid)
-
-            # 知识点标签统计 top 10（横向条形图）
-            tag_stats = crud.get_knowledge_stats(db, subject=subject, limit=10, user_id=uid)
-
-            # 知识点 × 掌握状态（堆叠柱状图）
-            tag_status_stats = crud.get_tag_status_stats(db, subject=subject, limit=10, user_id=uid)
-
-            # 知识点 × 题型（热力图）
-            tag_type_stats = crud.get_tag_type_stats(db, subject=subject, tag_limit=8, user_id=uid)
-
-            # 最近30天每日新增 + 已掌握趋势
-            daily_counts = crud.get_daily_counts(db, days=30, subject=subject, user_id=uid)
-
-            return jsonify({
-                'success': True,
-                'subjects': subjects,
-                'review_stats': review_stats,
-                'total_questions': statistics['total_questions'],
-                'by_subject': statistics['by_subject'],
-                'tag_stats': tag_stats,
-                'tag_status_stats': tag_status_stats,
-                'tag_type_stats': tag_type_stats,
-                'daily_counts': daily_counts,
-            })
-
-    except Exception as e:
-        logger.exception("获取 Dashboard 统计失败")
-        return jsonify({'success': False, 'error': '获取统计失败，请稍后重试'}), 500
-
-
-@app.route('/api/save-to-db', methods=['POST'])
-def save_to_db():
-    """
-    将分割好的题目导入错题库（仅入库，不导出 Markdown）
-
-    Request Body:
-        {
-            "selected_ids": ["q_0", "q_1", ...]   # 选中的题目 ID 列表
-        }
-    """
-    try:
-        data = request.get_json(silent=True) or {}
-        selected_ids = data.get('selected_ids', [])
-
-        if not isinstance(selected_ids, list) or not selected_ids:
-            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
-
-        results_dir = settings.results_dir
-        questions_file = os.path.join(results_dir, "questions.json")
-        if not os.path.exists(questions_file):
-            return jsonify({'success': False, 'error': '请先分割题目'}), 400
-
-        with open(questions_file, 'r', encoding='utf-8') as f:
-            questions = json.load(f)
-
-        selected_questions = [q for q in questions if q.get('question_id') in selected_ids]
-        if not selected_questions:
-            return jsonify({'success': False, 'error': '未找到选中的题目'}), 400
-
-        # 合并前端传来的 answer/user_answer（按 question_id 匹配）
-        answers_map = {a['question_id']: a for a in data.get('answers', []) if 'question_id' in a}
-        for sq in selected_questions:
-            qid = sq.get('question_id')
-            if qid and qid in answers_map:
-                if 'answer' in answers_map[qid]:
-                    sq['answer'] = answers_map[qid]['answer']
-                if 'user_answer' in answers_map[qid]:
-                    sq['user_answer'] = answers_map[qid]['user_answer']
-
-        # 读取科目信息
-        subject = _read_split_subject()
-
-        with session_lock:
-            batch_info = {
-                "original_filename": ", ".join(
-                    session_files.get(k, {}).get("filename", "未知")
-                    for k in session_file_order
-                ),
-                "subject": subject,
-                "file_path": "",
-            }
-
-        with SessionLocal() as db:
-            result = crud.save_questions_to_db(db, selected_questions, batch_info, user_id=session.get('user_id'))
-
-        return jsonify({
-            'success': True,
-            'message': f'已导入 {result["created"]} 道题目（跳过 {result["duplicates"]} 道重复）',
-            'created': result['created'],
-            'duplicates': result['duplicates'],
-        })
-
-    except Exception as e:
-        logger.exception("导入错题库失败")
-        return jsonify({'success': False, 'error': '导入错题库失败，请稍后重试'}), 500
-
-
-@app.route('/api/ai-analysis', methods=['POST'])
-def ai_analysis():
-    """
-    AI 错题分析（骨架路由）
-
-    接收一组题目 ID，调用 AI 对这些错题进行综合分析，返回：
-    - 薄弱知识点诊断
-    - 错因归纳
-    - 针对性复习建议
-
-    Request Body:
-        {
-            "question_ids": [1, 2, 3]   # 必填，至少 1 道，最多 20 道
-        }
-
-    Response:
-        {
-            "success": true,
-            "analysis": {
-                "summary": "综合诊断摘要",
-                "weak_points": ["知识点A", "知识点B"],
-                "error_patterns": [
-                    {"pattern": "错因类型", "description": "详细描述", "question_ids": [1, 2]}
-                ],
-                "suggestions": ["建议1", "建议2"],
-                "per_question": [
-                    {"question_id": 1, "diagnosis": "该题错因分析", "hint": "解题提示"}
-                ]
-            }
-        }
-
-    TODO（后端开发者）:
-        1. 从数据库查询题目详情（content_json, options_json, user_answer, knowledge_tags）
-        2. 构建 prompt，将题目内容 + 用户答案 + 知识点标签传给 LLM
-        3. 调用 LLM（建议使用 llm.py 中的 init_model，支持 openai/anthropic）
-        4. 解析 LLM 返回的结构化结果，填充 analysis 字段
-        5. 可选：将分析结果缓存到数据库，避免重复分析
-    """
-    try:
-        data = request.get_json(silent=True) or {}
-        question_ids = data.get('question_ids', [])
-
-        if not isinstance(question_ids, list) or not question_ids:
-            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
-
-        if len(question_ids) > 20:
-            return jsonify({'success': False, 'error': '单次最多分析 20 道题目'}), 400
-
-        with SessionLocal() as db:
-            questions = crud.get_questions_by_ids(db, question_ids)
-            if not questions:
-                return jsonify({'success': False, 'error': '未找到对应题目'}), 404
-
-            # ── TODO: 替换为真实 AI 分析逻辑 ──────────────────
-            # 当前返回占位数据，供前端联调使用
-            knowledge_tags = set()
-            per_question = []
-            for q in questions:
-                tags = [m.tag.tag_name for m in (q.tags or []) if m.tag]
-                knowledge_tags.update(tags)
-                per_question.append({
-                    'question_id': q.id,
-                    'diagnosis': f'题目 #{q.id} 的错因分析（待 AI 生成）',
-                    'hint': f'题目 #{q.id} 的解题提示（待 AI 生成）',
-                })
-
-            analysis = {
-                'summary': f'已选择 {len(questions)} 道题目进行分析（AI 分析功能待接入）',
-                'weak_points': list(knowledge_tags) or ['暂无知识点数据'],
-                'error_patterns': [
-                    {
-                        'pattern': '待 AI 识别',
-                        'description': '错因模式将由 AI 自动归纳',
-                        'question_ids': [q.id for q in questions],
-                    }
-                ],
-                'suggestions': [
-                    '建议回顾相关知识点章节',
-                    '建议针对薄弱环节进行专项练习',
-                ],
-                'per_question': per_question,
-            }
-            # ── END TODO ──────────────────────────────────────
-
-            return jsonify({
-                'success': True,
-                'analysis': analysis,
-            })
-
-    except Exception as e:
-        logger.exception("AI 分析失败")
-        return jsonify({'success': False, 'error': 'AI 分析失败，请稍后重试'}), 500
-
-
-@app.route('/api/export-from-db', methods=['POST'])
-def export_from_db():
-    """从数据库按 ID 列表导出 Markdown 错题本"""
-    try:
-        data = request.get_json(silent=True) or {}
-        selected_ids = data.get('selected_ids', [])
-
-        if not isinstance(selected_ids, list) or not selected_ids:
-            return jsonify({'success': False, 'error': '请选择至少一道题目'}), 400
-
-        with SessionLocal() as db:
-            questions_orm = crud.get_questions_by_ids(db, selected_ids)
-            if not questions_orm:
-                return jsonify({'success': False, 'error': '未找到对应题目'}), 404
-
-            # 转换为 export_wrongbook_md 所需的 dict 格式
-            questions = []
-            for q in questions_orm:
-                questions.append({
-                    'question_id': str(q.id),
-                    'question_type': q.question_type,
-                    'content_blocks': json.loads(q.content_json) if q.content_json else [],
-                    'options': json.loads(q.options_json) if q.options_json else None,
-                    'has_formula': q.has_formula,
-                    'has_image': q.has_image,
-                    'knowledge_tags': [m.tag.tag_name for m in (q.tags or []) if m.tag],
-                })
-
-            str_ids = [str(qid) for qid in selected_ids]
-            output_path = export_wrongbook_md(questions, str_ids)
-
-        return jsonify({
-            'success': True,
-            'message': '错题本导出成功',
-            'output_path': output_path,
-        })
-
-    except Exception as e:
-        logger.exception("从数据库导出失败")
-        return jsonify({'success': False, 'error': '导出失败，请稍后重试'}), 500
-
-
-# ============================================================
-# 教学辅导对话 API
-# ============================================================
-
-
-@app.route('/api/question/<int:question_id>/answer', methods=['PUT'])
-def save_question_answer(question_id):
-    """保存题目答案（Markdown 格式，用于 AI 辅导）"""
-    try:
-        data = request.get_json(silent=True) or {}
-        answer = data.get('answer')
-        if answer is None:
-            return jsonify({'success': False, 'error': '缺少 answer 字段'}), 400
-        if len(answer) > 50000:
-            return jsonify({'success': False, 'error': '答案内容过长（最多 50000 字符）'}), 400
-
-        with SessionLocal() as db:
-            question = crud.update_question_answer(db, question_id, answer)
-            if not question:
-                return jsonify({'success': False, 'error': '题目不存在'}), 404
-
-            return jsonify({
-                'success': True,
-                'message': '答案已保存',
-                'answer': question.answer,
-            })
-
-    except Exception as e:
-        logger.exception("保存答案失败")
-        return jsonify({'success': False, 'error': '保存答案失败，请稍后重试'}), 500
-
-
-@app.route('/api/question/<int:question_id>/chats', methods=['GET'])
-def get_question_chats(question_id):
-    """获取某道题目的所有对话会话"""
-    try:
-        with SessionLocal() as db:
-            sessions = crud.get_chat_sessions_by_question(db, question_id)
-            return jsonify({
-                'success': True,
-                'sessions': [_serialize_chat_session(s) for s in sessions],
-            })
-    except Exception as e:
-        logger.exception("获取对话列表失败")
-        return jsonify({'success': False, 'error': '获取对话列表失败'}), 500
-
-
-@app.route('/api/chat', methods=['POST'])
-def create_chat():
-    """创建新对话"""
-    try:
-        data = request.get_json(silent=True) or {}
-        question_id = data.get('question_id')
-        if not question_id:
-            return jsonify({'success': False, 'error': '缺少 question_id'}), 400
-
-        with SessionLocal() as db:
-            question = db.query(Question).filter(Question.id == question_id).first()
-            if not question:
-                return jsonify({'success': False, 'error': '题目不存在'}), 404
-
-            session = crud.create_chat_session(db, question_id)
-            return jsonify({
-                'success': True,
-                'session': _serialize_chat_session(session),
-            })
-
-    except Exception as e:
-        logger.exception("创建对话失败")
-        return jsonify({'success': False, 'error': '创建对话失败'}), 500
-
-
-@app.route('/api/chat/sessions', methods=['GET'])
-def get_chat_sessions():
-    """分页获取所有对话会话"""
-    try:
-        page = max(1, request.args.get('page', 1, type=int))
-        page_size = min(100, max(1, request.args.get('page_size', 20, type=int)))
-
-        with SessionLocal() as db:
-            sessions, total = crud.get_all_chat_sessions(db, page=page, page_size=page_size)
-            total_pages = (total + page_size - 1) // page_size
-
-            return jsonify({
-                'success': True,
-                'sessions': [_serialize_chat_session(s) for s in sessions],
-                'total': total,
-                'page': page,
-                'total_pages': total_pages,
-            })
-
-    except Exception as e:
-        logger.exception("获取对话会话列表失败")
-        return jsonify({'success': False, 'error': '获取对话会话列表失败'}), 500
-
-
-@app.route('/api/chat/<int:session_id>/messages', methods=['GET'])
-def get_chat_messages(session_id):
-    """游标分页获取对话消息"""
-    try:
-        limit = min(100, max(1, request.args.get('limit', 30, type=int)))
-        before_id = request.args.get('before_id', type=int)
-
-        with SessionLocal() as db:
-            result = crud.get_chat_messages(db, session_id, limit=limit, before_id=before_id)
-            return jsonify({
-                'success': True,
-                'messages': result['messages'],
-                'hasMore': result['hasMore'],
-            })
-
-    except Exception as e:
-        logger.exception("获取对话消息失败")
-        return jsonify({'success': False, 'error': '获取对话消息失败'}), 500
-
-
-@app.route('/api/chat/<int:session_id>/stream', methods=['POST'])
-def stream_chat(session_id):
-    """SSE 流式对话"""
-    from flask import Response
-    from teach_agent import stream_teach
-
-    try:
-        data = request.get_json(silent=True) or {}
-        message = data.get('message', '').strip()
-        model_provider = data.get('model_provider', 'openai')
-        model_name = data.get('model_name') or None
-
-        if not message:
-            return jsonify({'success': False, 'error': '消息不能为空'}), 400
-
-        if not settings.is_valid_provider(model_provider):
-            return jsonify({'success': False, 'error': f'不支持的模型供应商: {model_provider}'}), 400
-
-        # 从数据库加载用户的 LLM 凭据
-        user_id = session.get('user_id')
-        if user_id:
-            settings.load_providers_from_db(user_id)
-
-        with SessionLocal() as db:
-            from db.models import ChatSession as ChatSessionModel, QuestionTagMapping
-            from sqlalchemy.orm import selectinload
-            chat_session = db.query(ChatSessionModel).options(
-                selectinload(ChatSessionModel.question)
-                .selectinload(Question.batch),
-                selectinload(ChatSessionModel.question)
-                .selectinload(Question.tags)
-                .selectinload(QuestionTagMapping.tag),
-            ).filter(ChatSessionModel.id == session_id).first()
-            if not chat_session:
-                return jsonify({'success': False, 'error': '对话不存在'}), 404
-
-            question = chat_session.question
-            if not question:
-                return jsonify({'success': False, 'error': '关联题目不存在'}), 404
-
-            # 序列化题目数据
-            q_data = _serialize_question(question)
-
-            # 加载历史消息（最近 20 条）
-            history_result = crud.get_chat_messages(db, session_id, limit=20)
-            history = [{"role": m["role"], "content": m["content"]} for m in history_result["messages"]]
-
-            # 追加用户消息
-            history.append({"role": "user", "content": message})
-            crud.add_chat_message(db, session_id, "user", message)
-
-        def generate():
-            full_response = []
-            try:
-                for token in stream_teach(
-                    question=q_data,
-                    messages=history,
-                    provider=model_provider,
-                    model_name=model_name,
-                ):
-                    full_response.append(token)
-                    yield f"data: {json.dumps({'token': token}, ensure_ascii=False)}\n\n"
-            except Exception as e:
-                logger.exception("流式对话错误")
-                yield f"data: {json.dumps({'error': str(e)}, ensure_ascii=False)}\n\n"
-
-            # 保存完整的 assistant 回复
-            assistant_content = "".join(full_response)
-            if assistant_content:
-                try:
-                    with SessionLocal() as db:
-                        crud.add_chat_message(db, session_id, "assistant", assistant_content)
-                except Exception as e:
-                    logger.error(f"保存 assistant 回复失败: {e}")
-
-            yield f"data: {json.dumps({'done': True})}\n\n"
-
-        resp = Response(generate(), mimetype='text/event-stream')
-        resp.headers['Cache-Control'] = 'no-cache'
-        resp.headers['X-Accel-Buffering'] = 'no'
-        return resp
-
-    except Exception as e:
-        logger.exception("流式对话失败")
-        return jsonify({'success': False, 'error': '对话失败，请稍后重试'}), 500
-
-
 if __name__ == '__main__':
-    # 确保运行时目录存在
+    # 创建运行时目录（uploads、pages、results 等）
     settings.ensure_dirs()
-    # 初始化数据库
+
+    # 初始化数据库（建表 + 自动迁移）
     init_db()
-    # 自动迁移（添加新列等）
     from db.migrate import migrate
     migrate()
-    print("[数据库] 初始化完成")
 
-    print("=" * 60)
-    print("错题本生成系统 - Web应用")
-    print("=" * 60)
-    print("访问地址: http://localhost:5001")
-    print("=" * 60)
+    logger.info("错题本生成系统 - API 服务")
+    logger.info("API 地址: http://localhost:5001")
+    logger.info("=" * 50)
 
+    # 从环境变量读取调试模式开关（默认关闭）
+    # 开启后 Flask 会自动重载代码变更，并在浏览器显示详细错误信息
+    debug = os.getenv('FLASK_DEBUG', 'false').lower() == 'true'
     app.run(
-        host='0.0.0.0', port=5001, debug=True,
-        exclude_patterns=["*site-packages*"],
+        host='0.0.0.0', port=5001, debug=debug,
+        threaded=True,                           # 多线程处理并发请求
+        exclude_patterns=["*site-packages*"],    # 排除第三方库文件变更触发的重载
     )
diff --git a/data-processing/README.md b/data-processing/README.md
deleted file mode 100644
index 31434c33..00000000
--- a/data-processing/README.md
+++ /dev/null
@@ -1,104 +0,0 @@
-# Data Processing
-
-OCR 图像预处理与识别准确率评估系统。
-
-## 目录结构
-
-```
-data-processing/
-├── preprocessing/              # 图像预处理模块
-│   ├── data_preprocessing.py   # 预处理核心
-│   ├── preprocessing_mode.json # 预处理几种方案
-│   └── README.md
-│   
-│
-├── processing/                    # 基于OmniDocBench的识别评估模块
-│   ├── evaluation_recognition.py  # 记录识别结果
-│   ├── exam_accuracy_results.json      # 试卷样例识别结果
-│   ├── jiaocai_accuracy_results.json   #教程样例识别结果
-│   ├── OmniDocBench.json         # OmniDocBench数据集答案
-│   └── sample_image              # 优秀样例图片
-│       ├──exam/                  # 试卷
-│       └──jiaocai                # 教材
-│
-├── standard_doc/                 # 图像扫描规范文档、数据标注规范
-│   ├── DATA_LABELING_STANDARDS.md # 数据标注
-│   └── SCANNING_STANDARDS.md      # 扫描规范
-└── README.md
-```
-
-## 快速开始
-
-### 1. 原始图片识别评估
-
-```bash
-cd processing
-python evaluation_recognition.py
-```
-
-评估原始图片的 OCR 识别准确率，与 OmniDocBench 标注对比，生成 `exam_accuracy_results.json`。
-
-### 2. 图像预处理
-
-```python
-from preprocessing.data_preprocessing import run_pipeline
-
-# 单张图片
-run_pipeline(file_path="test.png", mode="auto", save_log=True)
-
-# 批量处理
-import glob
-for img in glob.glob("sample_image/exam/*.png"):
-    run_pipeline(file_path=img, mode="auto", save_log=True)
-```
-
-**预处理模式**：
-
-| 模式 | 适用场景 |
-|------|---------|
-| `auto` | 自动选择最佳策略（推荐） |
-| `standard` | 一般文档 |
-| `lightweight` | 高质量文档、数学公式密集 |
-| `illumination_rescue` | 光照不均、阴影、反光 |
-| `screen_photography` | 屏幕翻拍、摩尔纹 |
-| `aggressive_compression` | 文件压缩 |
-
-
-```
-
-## 实验结果
-
-基于 OmniDocBench 数据集：
-
-- **原始平均准确率**: 83.97%
-- **预处理后平均准确率**: 85.50%
-- **平均提升**: 1.5-2%
-- **中文试卷**: illumination_rescue 模式提升 2-7%
-- **英文数学试卷**: lightweight 模式保持原有准确率
-
-## 配置
-
-### 环境变量
-
-创建 `.env` 文件：
-
-```env
-PADDLEOCR_API_URL=https://your-api-endpoint
-PADDLEOCR_API_TOKEN=your-token-here
-```
-
-### 自定义预处理参数
-
-编辑 `preprocessing/preprocessing_mode.json` 添加自定义模式。
-
-## 注意事项
-
-1. **API 限流**: 建议 `MAX_WORKERS` 不超过 5
-2. **文件命名**: 图片文件名需与 `OmniDocBench.json` 中的 `image_path` 匹配
-3. **预处理策略**: 数学公式密集文档避免使用 `illumination_rescue`
-
-## 相关文档
-
-- [预处理模块详细说明](preprocessing/README.md)
-- [数据标注规范](standard_doc/DATA_LABELING_STANDARDS.md)
-- [扫描规范](standard_doc/SCANNING_STANDARDS.md)
diff --git a/data-processing/preprocessing/README.md b/data-processing/preprocessing/README.md
deleted file mode 100644
index 69702fb6..00000000
--- a/data-processing/preprocessing/README.md
+++ /dev/null
@@ -1,101 +0,0 @@
-# 图像预处理模块
-
-智能图像预处理系统，用于优化 OCR 识别效果。
-
-## 核心文件
-
-- `data_preprocessing.py` - 主预处理程序
-- `preprocessing_mode.json` - 预处理策略配置
-
-
-## 快速开始
-
-### 1. 单张图片预处理
-
-```python
-from data_preprocessing import run_pipeline
-
-# 自动模式（推荐）
-run_pipeline(
-    file_path="your_image.png",
-    mode="auto",  # 自动选择最佳策略
-    save_log=True
-)
-
-# 手动指定模式
-run_pipeline(
-    file_path="your_image.png",
-    mode="lightweight",  # 可选: standard, lightweight, illumination_rescue, screen_photography, aggressive_compression
-    save_log=True
-)
-```
-
-### 2. 批量处理
-
-```python
-import glob
-from data_preprocessing import run_pipeline
-
-image_files = glob.glob("your_folder/*.png")
-for img_path in image_files:
-    run_pipeline(file_path=img_path, mode="auto", save_log=True)
-```
-
-
-
-## 预处理模式说明
-
-| 模式 | 适用场景 | 特点 |
-|------|---------|------|
-| `auto` | 所有场景（推荐） | 自动诊断并选择最佳策略 |
-| `standard` | 一般文档 | 平衡速度与精度 |
-| `lightweight` | 高质量文档、数学公式密集 | 仅做基础优化，保护细节 |
-| `illumination_rescue` | 光照不均、阴影、反光 | 对比度增强，光照修复 |
-| `screen_photography` | 屏幕翻拍、摩尔纹 | 去除摩尔纹 |
-| `aggressive_compression` | 网络差、文件大 | 极限压缩 |
-
-## Auto 模式智能路由
-
-系统会自动分析图像特征并选择最佳策略：
-
-- **数学公式密集 / 高质量文档** → `lightweight`
-- **光照问题明显** → `illumination_rescue`
-- **屏幕翻拍特征** → `screen_photography`
-- **大文件低文本密度** → `aggressive_compression`
-- **其他** → `standard`
-
-## 输出结构
-
-```
-output/
-├── image_name_processed_strategy/
-│   └── api_analysis.log          # API 识别结果
-processed_exam_images/
-└── image_name.png                 # 预处理后的图片
-```
-
-## 配置自定义
-
-编辑 `preprocessing_mode.json` 可自定义预处理参数：
-
-```json
-{
-  "modes": {
-    "your_custom_mode": {
-      "description": "自定义模式说明",
-      "target_dpi": 150,
-      "max_side_len": 1920,
-      "jpeg_quality": 95,
-      "enable_deskew": true,
-      "enable_contrast_enhance": false,
-      "enable_moire_removal": false
-    }
-  }
-}
-```
-
-## 性能优化建议
-
-1. **中文试卷**：auto 模式会自动选择 `illumination_rescue`，提升 2-7%
-2. **英文数学试卷**：auto 模式会自动选择 `lightweight`，避免符号识别下降
-3. **批量处理**：建议使用 auto 模式，系统会为每张图片选择最优策略
diff --git a/data-processing/preprocessing/data_preprocessing.py b/data-processing/preprocessing/data_preprocessing.py
deleted file mode 100644
index 4d72dc6e..00000000
--- a/data-processing/preprocessing/data_preprocessing.py
+++ /dev/null
@@ -1,508 +0,0 @@
-import base64
-import os
-import requests
-import time
-import json
-import math
-import cv2
-import numpy as np
-from typing import Dict, Any
-from dotenv import load_dotenv
-from PIL import Image
-
-# ==========================================
-# 全局配置参数
-# ==========================================
-load_dotenv()  # 从 .env 文件加载环境变量
-API_URL = os.getenv("PADDLEOCR_API_URL")  
-TOKEN = os.getenv("PADDLEOCR_API_TOKEN")  
-
-OUTPUT_DIR = "output"                 # API 识别结果、标记图片和日志保存目录
-PROCESSED_DIR = "processed_exam_images"    # 预处理后的中间图片保存目录
-CONFIG_FILE = "preprocessing_mode.json" # 预处理配置文件路径
-
-# ==========================================
-# 模块一：配置加载
-# ==========================================
-def load_preprocessing_config(mode_name=None):
-    """加载预处理策略配置"""
-    if not os.path.exists(CONFIG_FILE):
-        raise FileNotFoundError(f"配置文件 {CONFIG_FILE} 不存在！请先创建。")
-        
-    with open(CONFIG_FILE, 'r', encoding='utf-8') as f:
-        config_data = json.load(f)
-        
-    if mode_name is None:
-        mode_name = config_data.get("default_mode", "standard")
-        
-    if mode_name not in config_data["modes"]:
-        raise ValueError(f"未找到预处理模式 '{mode_name}'，请检查配置文件。")
-        
-    print(f"[Config] 已加载预处理模式: '{mode_name}'")
-    print(f"[Config] 策略描述: {config_data['modes'][mode_name].get('description', '')}")
-    return config_data["modes"][mode_name]
-
-# ==========================================
-# 模块1.5：图像特征分析 (自适应诊断)
-# ==========================================
-def get_file_size_kb(path: str) -> float:
-    return round(os.path.getsize(path) / 1024.0, 2)
-
-def get_dpi(path: str, default: int = 72) -> int:
-    try:
-        with Image.open(path) as im:
-            dpi = im.info.get('dpi')
-            if dpi and isinstance(dpi, tuple):
-                return int(dpi[0])
-    except Exception:
-        pass
-    return default
-
-def to_gray(img: np.ndarray) -> np.ndarray:
-    if len(img.shape) == 3:
-        return cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-    return img
-
-def estimate_brightness_and_contrast(gray: np.ndarray) -> Dict[str, float]:
-    arr = gray.astype(np.float32)
-    mean = float(np.mean(arr))
-    std = float(np.std(arr))
-    return {"brightness": round(mean, 2), "contrast": round(std, 2)}
-
-def estimate_sharpness(gray: np.ndarray) -> float:
-    # Variance of Laplacian (常用的清晰度测量)
-    var_lap = cv2.Laplacian(gray, cv2.CV_64F).var()
-    return float(var_lap)
-
-def estimate_noise(img: np.ndarray) -> float:
-    # 通过原图与高斯模糊图之差估计高频噪声
-    blur = cv2.GaussianBlur(img, (3, 3), 0)
-    if img.ndim == 3:
-        diff = cv2.cvtColor(cv2.absdiff(img, blur), cv2.COLOR_BGR2GRAY)
-    else:
-        diff = cv2.absdiff(img, blur)
-    return float(np.std(diff))
-
-def estimate_edge_density(gray: np.ndarray) -> float:
-    edges = cv2.Canny(gray, 100, 200)
-    edge_ratio = float(np.count_nonzero(edges)) / (gray.shape[0] * gray.shape[1])
-    return round(edge_ratio, 4)
-
-def estimate_text_density(gray: np.ndarray) -> float:
-    # 自适应二值化后计算黑色像素比例，作为文本/线条密度的近似
-    th = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
-                               cv2.THRESH_BINARY_INV, 25, 15)
-    density = float(np.count_nonzero(th)) / (gray.shape[0] * gray.shape[1])
-    return round(density, 4)
-
-def detect_color_mode(img: np.ndarray) -> str:
-    if len(img.shape) == 2:
-        return "grayscale"
-    # 如果三通道间差异很小，认为是灰度图
-    channel_diff = np.max(img.astype(np.int32) - img[..., 0:1].astype(np.int32))
-    if channel_diff < 5:
-        return "grayscale"
-    return "color"
-
-def estimate_skew_angle(gray: np.ndarray) -> float:
-    # 使用 Hough 直线检测（在边缘图上）来估计主方向角度
-    h, w = gray.shape[:2]
-    # Resize reduce cost
-    scale = 600.0 / max(h, w) if max(h, w) > 600 else 1.0
-    small = cv2.resize(gray, (int(w * scale), int(h * scale)), interpolation=cv2.INTER_AREA)
-    edges = cv2.Canny(small, 50, 150)
-    lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=50, minLineLength=30, maxLineGap=10)
-    angles = []
-    if lines is None:
-        return 0.0
-    for l in lines:
-        x1, y1, x2, y2 = l[0]
-        angle = math.degrees(math.atan2(y2 - y1, x2 - x1))
-        # 只考虑接近水平的直线，过滤垂直线
-        if abs(angle) < 45:
-            angles.append(angle)
-    if not angles:
-        return 0.0
-    # 取中位数以减少离群点影响
-    med = float(np.median(np.array(angles)))
-    # 返回图片坐标系下的倾斜角（度），正值表示逆时针旋转
-    return round(med / scale if scale != 1.0 else med, 2)
-
-def analyze_image(path: str) -> Dict[str, Any]:
-    if not os.path.exists(path):
-        raise FileNotFoundError(path)
-
-    pil_img = Image.open(path)
-    w, h = pil_img.size
-    file_size_kb = get_file_size_kb(path)
-    dpi = get_dpi(path)
-
-    # Load as OpenCV image
-    img_cv = cv2.imdecode(np.fromfile(path, dtype=np.uint8), cv2.IMREAD_UNCHANGED)
-    if img_cv is None:
-        # fallback via PIL
-        img_cv = cv2.cvtColor(np.array(pil_img.convert('RGB')), cv2.COLOR_RGB2BGR)
-
-    gray = to_gray(img_cv)
-
-    bc = estimate_brightness_and_contrast(gray)
-    sharpness = estimate_sharpness(gray)
-    noise = estimate_noise(img_cv)
-    edge_density = estimate_edge_density(gray)
-    text_density = estimate_text_density(gray)
-    color_mode = detect_color_mode(img_cv)
-    skew = estimate_skew_angle(gray)
-
-    aspect_ratio = round(w / h, 4) if h != 0 else None
-
-    return {
-        "file": os.path.basename(path),
-        "path": path,
-        "file_size_kb": file_size_kb,
-        "dpi": dpi,
-        "resolution": f"{w}x{h}",
-        "aspect_ratio": aspect_ratio,
-        "color_mode": color_mode,
-        "brightness": bc["brightness"],
-        "contrast": bc["contrast"],
-        "sharpness_variance_laplacian": round(sharpness, 2),
-        "noise_stddev": round(noise, 2),
-        "edge_density": edge_density,
-        "text_density": text_density,
-        "skew_angle_degrees": skew,
-        "is_blurry": sharpness < 100.0
-    }
-
-# ==========================================
-# 模块二：图像 DPI 读取与真实重采样
-# ==========================================
-def get_image_dpi(file_path):
-    """读取图像文件的 DPI 元数据标签"""
-    try:
-        with Image.open(file_path) as img:
-            dpi = img.info.get('dpi')
-            if dpi:
-                return f"{int(dpi[0])} x {int(dpi[1])}"
-            else:
-                return "Not Set (System Default)"
-    except Exception as e:
-        return f"Read Failed: {e}"
-
-def resample_image_by_dpi(image_path, target_dpi=150, default_original_dpi=72):
-    """真正的 DPI 转换器：不仅修改标签，还会物理缩放像素矩阵。"""
-    with Image.open(image_path) as img:
-        orig_dpi_tuple = img.info.get('dpi')
-        original_dpi = orig_dpi_tuple[0] if orig_dpi_tuple else default_original_dpi
-            
-        scale_factor = target_dpi / original_dpi
-        
-        if scale_factor == 1.0:
-            return cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
-            
-        new_width = int(img.width * scale_factor)
-        new_height = int(img.height * scale_factor)
-        
-        print(f"[Preprocess] DPI Resampling: {original_dpi} -> {target_dpi} DPI")
-        
-        # 执行高质量重采样 (Lanczos)
-        resized_img = img.resize((new_width, new_height), Image.Resampling.LANCZOS)
-        return cv2.cvtColor(np.array(resized_img), cv2.COLOR_RGB2BGR)
-
-# ==========================================
-# 模块三：核心 CV 预处理算法
-# ==========================================
-def resize_image_for_paddle(image, max_side_len=1920):
-    """限制图像的最大边长"""
-    h, w = image.shape[:2]
-    if max(h, w) > max_side_len:
-        if h > w:
-            new_h = max_side_len
-            new_w = int(w * (max_side_len / h))
-        else:
-            new_w = max_side_len
-            new_h = int(h * (max_side_len / w))
-        image = cv2.resize(image, (new_w, new_h), interpolation=cv2.INTER_AREA)
-        print(f"[Preprocess] Size Truncation: ({w}x{h}) -> ({new_w}x{new_h})")
-    return image
-
-def deskew_image(image):
-    """倾斜校正"""
-    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
-    edges = cv2.Canny(gray, 50, 150, apertureSize=3)
-    lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 100, minLineLength=100, maxLineGap=10)
-    
-    if lines is not None:
-        angles = []
-        for line in lines:
-            x1, y1, x2, y2 = line[0]
-            angle = np.degrees(np.arctan2(y2 - y1, x2 - x1))
-            if -45 < angle < 45:
-                angles.append(angle)
-                
-        if angles:
-            median_angle = np.median(angles)
-            if abs(median_angle) > 0.5: 
-                h, w = image.shape[:2]
-                center = (w // 2, h // 2)
-                M = cv2.getRotationMatrix2D(center, median_angle, 1.0)
-                rotated = cv2.warpAffine(image, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_CONSTANT, borderValue=(255, 255, 255))
-                print(f"[Preprocess] Deskew applied: {median_angle:.2f} degrees")
-                return rotated
-    return image
-
-def enhance_contrast_clahe(image, clip_limit=2.0, tile_size=8):
-    """CLAHE 增强图像局部对比度"""
-    lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)
-    l, a, b = cv2.split(lab)
-    clahe = cv2.createCLAHE(clipLimit=clip_limit, tileGridSize=(tile_size, tile_size))
-    cl = clahe.apply(l)
-    limg = cv2.merge((cl, a, b))
-    final = cv2.cvtColor(limg, cv2.COLOR_LAB2BGR)
-    print(f"[Preprocess] Contrast enhanced (CLAHE: clip={clip_limit}, tile={tile_size})")
-    return final
-
-def remove_moire_pattern(image):
-    """高斯滤波弱化摩尔纹"""
-    blurred = cv2.GaussianBlur(image, (3, 3), 0)
-    print("[Preprocess] Moire pattern removed (Gaussian Blur)")
-    return blurred
-
-# ==========================================
-# 模块四：基于配置的预处理调度器
-# ==========================================
-def preprocess_image(input_path, config):
-    """根据传入的配置字典执行预处理流水线"""
-    os.makedirs(PROCESSED_DIR, exist_ok=True)
-    base_name = os.path.basename(input_path)
-    file_name_without_ext = os.path.splitext(base_name)[0]
-    file_ext = os.path.splitext(base_name)[1]
-
-    # 保存为原始文件名（保持原扩展名或使用 jpg）
-    if file_ext.lower() in ['.jpg', '.jpeg', '.png', '.bmp', '.webp']:
-        output_path = os.path.join(PROCESSED_DIR, base_name)
-    else:
-        output_path = os.path.join(PROCESSED_DIR, f"{file_name_without_ext}.jpg")
-    
-    print(f"\n[Step 1/3] Local Preprocessing: {input_path}")
-    
-    try:
-        image = resample_image_by_dpi(input_path, target_dpi=config.get('target_dpi', 150))
-    except Exception as e:
-        raise ValueError(f"Image read/resample failed: {e}")
-
-    # 根据配置执行流水线
-    image = resize_image_for_paddle(image, max_side_len=config.get('max_side_len', 1920))
-    
-    if config.get('enable_deskew', False):
-        image = deskew_image(image)
-        
-    if config.get('enable_moire_removal', False):
-        image = remove_moire_pattern(image)
-        
-    if config.get('enable_contrast_enhance', False):
-        clip_limit = config.get('clahe_clip_limit', 2.0)
-        tile_size = config.get('clahe_tile_size', 8)
-        image = enhance_contrast_clahe(image, clip_limit=clip_limit, tile_size=tile_size)
-
-    # 统一转换色彩空间并使用 PIL 保存，应用 jpeg_quality 压缩参数
-    image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
-    pil_img = Image.fromarray(image_rgb)
-    
-    target_dpi = config.get('target_dpi', 150)
-    jpeg_quality = config.get('jpeg_quality', 95)
-    
-    pil_img.save(output_path, format="JPEG", quality=jpeg_quality, dpi=(target_dpi, target_dpi))
-    
-    file_size_kb = os.path.getsize(output_path) / 1024
-    print(f"[Preprocess] Completed! Saved to: {output_path} (Size: {file_size_kb:.2f} KB, Quality: {jpeg_quality})")
-    return output_path
-
-# ==========================================
-# 模块五：云端 API 调用与结果解析
-# ==========================================
-def get_file_type(file_path):
-    ext = os.path.splitext(file_path)[-1].lower()
-    if ext == ".pdf": return 0
-    elif ext in [".jpg", ".jpeg", ".png", ".bmp", ".webp"]: return 1
-    else: raise ValueError(f"Unsupported file format: {ext}")
-
-def call_paddleocr_api(file_path):
-    print(f"\n[Step 2/3] Calling Cloud API...")
-    file_type = get_file_type(file_path)
-    with open(file_path, "rb") as file:
-        file_bytes = file.read()
-        file_data = base64.b64encode(file_bytes).decode("ascii")
-
-    headers = {"Authorization": f"token {TOKEN}", "Content-Type": "application/json"}
-    payload = {
-        "file": file_data,
-        "fileType": file_type, 
-        "useDocOrientationClassify": False,
-        "useDocUnwarping": False,
-        "useChartRecognition": False,
-    }
-
-    start_time = time.perf_counter()
-    response = requests.post(API_URL, json=payload, headers=headers)
-    response.raise_for_status()
-    api_latency = time.perf_counter() - start_time
-    
-    return response.json(), api_latency, response.json()
-
-def analyze_and_save(original_file_path, api_target_file, light_data, api_latency, raw_data, suffix="", save_log=False):
-    print(f"\n[Step 3/3] Parsing response...")
-    base_name = os.path.splitext(os.path.basename(original_file_path))[0]
-    test_output_dir = os.path.join(OUTPUT_DIR, f"{base_name}_{suffix}" if suffix else base_name)
-    os.makedirs(test_output_dir, exist_ok=True)
-
-    if save_log:
-        log_file_path = os.path.join(test_output_dir, "api_analysis.log")
-        with open(log_file_path, "w", encoding="utf-8") as log_file:
-            json.dump(light_data, log_file, ensure_ascii=False, indent=2)
-        print(f"[Preprocess] API info logged to {log_file_path}")
-
-    try:
-        res = light_data.get('result', {})
-        layout_results = res.get('layoutParsingResults', [{}])[0]
-        width, height = res.get('dataInfo', {}).get('width', 0), res.get('dataInfo', {}).get('height', 0)
-        num_blocks = len(layout_results.get('prunedResult', {}).get('parsing_res_list', []))
-        
-        text_content = layout_results.get('markdown', {}).get('text', '')
-        text_length = len(text_content)
-        actual_dpi = get_image_dpi(api_target_file) if get_file_type(api_target_file) == 1 else "PDF"
-
-        print("-" * 50)
-        print(f"[{suffix.upper()}] API Data Metrics")
-        print("-" * 50)
-        print(f"Latency     : {api_latency:.2f} s")
-        print(f"DPI Tag     : {actual_dpi}")
-        print(f"Resolution  : {width} x {height}")
-        print(f"Blocks      : {num_blocks}")
-        print(f"Text Length : {text_length} chars")
-        print("-" * 50)
-    except Exception:
-        pass 
-
-# ==========================================
-# 最终运行入口
-# ==========================================
-def run_pipeline(file_path, mode="auto", save_log=False):
-    """
-    端到端总控调度
-    :param file_path: 目标文件路径
-    :param mode: "auto" 会触发自动图像诊断路由，也可直接指定配置文件里的模式
-    """
-    print(f"\n{'='*60}")
-    print(f"Task Started: {file_path} | Requested Mode: {mode}")
-    print(f"{'='*60}")
-    
-    if not os.path.exists(file_path):
-        print(f"[Error] File not found: {file_path}")
-        return
-
-    is_image = (get_file_type(file_path) == 1)
-    diagnosis_info = None
-
-    # Step 0: 如果是图片，我们先进行物理体检（改动最小的解耦融合方案）
-    if is_image:
-        print("\n[Step 0/3] Image Auto-Diagnosis...")
-        try:
-            diagnosis_info = analyze_image(file_path)
-            print(f"  -> Features: Brightness={diagnosis_info['brightness']}, Blur={diagnosis_info['is_blurry']}, Skew={diagnosis_info['skew_angle_degrees']}")
-
-            # 核心融合点：如果开启了 Auto 模式，则利用诊断结果决定该用哪套 JSON 预设
-            if mode == "auto":
-                # 判断是否为数学公式密集文档（高文本密度 + 高边缘密度 + 较高清晰度）
-                is_math_heavy = (
-                    diagnosis_info['text_density'] > 0.08 and
-                    diagnosis_info['edge_density'] > 0.05 and
-                    diagnosis_info['sharpness_variance_laplacian'] > 150
-                )
-
-                # 判断是否为质量已经较好的文档
-                is_good_quality = (
-                    diagnosis_info['brightness'] >= 120 and
-                    diagnosis_info['brightness'] <= 200 and
-                    diagnosis_info['contrast'] >= 50 and
-                    not diagnosis_info['is_blurry'] and
-                    abs(diagnosis_info['skew_angle_degrees']) < 2.0
-                )
-
-                # 智能路由逻辑
-                if is_math_heavy or is_good_quality:
-                    # 数学公式密集或质量已经很好的文档，使用轻量级模式
-                    mode = "lightweight"
-                    print(f"  -> [Auto-Routing] Detected high-quality or math-heavy document, using lightweight mode")
-                elif diagnosis_info['brightness'] < 100 or diagnosis_info['brightness'] > 220 or diagnosis_info['contrast'] < 40:
-                    # 光照问题明显，使用光照拯救模式
-                    mode = "illumination_rescue"
-                    print(f"  -> [Auto-Routing] Detected illumination issues, using illumination_rescue mode")
-                elif diagnosis_info['noise_stddev'] > 3.0 and not diagnosis_info['is_blurry']:
-                    # 噪声明显但不模糊，可能是屏幕翻拍
-                    mode = "screen_photography"
-                    print(f"  -> [Auto-Routing] Detected screen photography patterns, using screen_photography mode")
-                elif diagnosis_info['file_size_kb'] > 3000 and diagnosis_info['text_density'] < 0.03:
-                    # 文件大但文本少，需要压缩
-                    mode = "aggressive_compression"
-                    print(f"  -> [Auto-Routing] Large file with low text density, using aggressive_compression mode")
-                else:
-                    # 默认使用标准模式
-                    mode = "standard"
-                    print(f"  -> [Auto-Routing] Using standard mode for general document")
-        except Exception as e:
-            print(f"  -> [Warning] Diagnosis failed: {e}")
-            if mode == "auto": mode = "standard"
-
-    if is_image and mode != "raw": 
-        # 读取配置并预处理
-        config = load_preprocessing_config(mode_name=mode)
-        api_target_file = preprocess_image(file_path, config)
-        suffix = f"processed_{mode}" 
-    else:
-        print(f"\n[Step 1/3] Skipping preprocessing, using raw file...")
-        api_target_file = file_path
-        suffix = "raw"
-        
-    try:
-        light_log, latency, raw_json = call_paddleocr_api(api_target_file)
-    except Exception as e:
-        print(f"[Error] API call failed: {e}")
-        return
-
-    # 把物理特征的诊断记录随同 API 结果一并写入 log 文件，辅助后期深度分析
-    if diagnosis_info and isinstance(light_log, dict):
-        light_log["image_diagnosis"] = diagnosis_info
-
-    analyze_and_save(file_path, api_target_file, light_log, latency, raw_json, suffix=suffix, save_log=save_log)
-    print("\n[Done] Pipeline execution completed successfully.")
-
-if __name__ == "__main__":
-    # ==========================================
-    # 1. 单张图片测试示例
-    # ==========================================
-    # test_file = r"test_image/raw_image/raw_test_1_copy.png"
-    
-    # # 使用自适应模式 "auto"，系统将自己进行体检并匹配预处理
-    # run_pipeline(
-    #     file_path=test_file, 
-    #     mode="standard",  # 可选: auto, standard, aggressive_compression, illumination_rescue, screen_photography, raw
-    #     save_log=True # 开启 Log 可以将诊断特征一并写入结果中
-    # )
-    
-    # ==========================================
-    # 2. 文件夹批处理测试示例 (若需使用，请取消下方注释)
-    # ==========================================
-    import glob
-    test_dir = r"D:\my_data\code\error_correction\data-processing\processing\sample_image\exam"
-    # 获取目录下所有的文件
-    image_files = glob.glob(os.path.join(test_dir, "*.*"))
-    
-    print(f"\n📁 启动文件夹批处理模式: {test_dir} (共找到 {len(image_files)} 个文件)")
-    for img_path in image_files:
-        # 仅处理常见图像格式
-        if img_path.lower().endswith(('.png', '.jpg', '.jpeg', '.bmp', '.webp')):
-            run_pipeline(
-                file_path=img_path, 
-                mode="auto",      # 批处理强烈建议用 auto 模式
-                save_log=True     # 批处理时建议开启日志以便后续分析
-            )
\ No newline at end of file
diff --git a/data-processing/preprocessing/examples.py b/data-processing/preprocessing/examples.py
deleted file mode 100644
index d7d44f34..00000000
--- a/data-processing/preprocessing/examples.py
+++ /dev/null
@@ -1,69 +0,0 @@
-"""
-预处理模块使用示例
-"""
-from data_preprocessing import run_pipeline
-import glob
-import os
-
-# ==========================================
-# 示例 1: 单张图片预处理（自动模式）
-# ==========================================
-def example_single_image_auto():
-    """单张图片自动预处理"""
-    test_file = r"test_image/sample.png"
-
-    if os.path.exists(test_file):
-        run_pipeline(
-            file_path=test_file,
-            mode="auto",  # 自动选择最佳策略
-            save_log=True
-        )
-
-
-# ==========================================
-# 示例 2: 单张图片预处理（指定模式）
-# ==========================================
-def example_single_image_manual():
-    """单张图片手动指定模式"""
-    test_file = r"test_image/sample.png"
-
-    if os.path.exists(test_file):
-        run_pipeline(
-            file_path=test_file,
-            mode="lightweight",  # 手动指定轻量级模式
-            save_log=True
-        )
-
-
-# ==========================================
-# 示例 3: 批量处理文件夹
-# ==========================================
-def example_batch_processing():
-    """批量处理文件夹中的所有图片"""
-    test_dir = r"test_image/batch"
-
-    # 获取所有图片文件
-    image_files = glob.glob(os.path.join(test_dir, "*.*"))
-
-    print(f"\n启动批处理: {test_dir} (共 {len(image_files)} 个文件)")
-
-    for img_path in image_files:
-        # 仅处理常见图像格式
-        if img_path.lower().endswith(('.png', '.jpg', '.jpeg', '.bmp', '.webp')):
-            print(f"\n处理: {os.path.basename(img_path)}")
-            run_pipeline(
-                file_path=img_path,
-                mode="auto",  # 批处理推荐使用 auto 模式
-                save_log=True
-            )
-
-
-
-if __name__ == "__main__":
-    # 取消注释以运行相应示例
-
-    # example_single_image_auto()
-    # example_single_image_manual()
-    # example_batch_processing()
-
-    print("请取消注释相应的示例函数以运行")
diff --git a/data-processing/preprocessing/preprocessing_mode.json b/data-processing/preprocessing/preprocessing_mode.json
deleted file mode 100644
index 23ff8765..00000000
--- a/data-processing/preprocessing/preprocessing_mode.json
+++ /dev/null
@@ -1,52 +0,0 @@
-{
-  "default_mode": "standard",
-  "modes": {
-    "standard": {
-      "description": "标准通用模式：平衡了速度与精度，适合大部分正常文档",
-      "target_dpi": 100,
-      "max_side_len": 1920,
-      "jpeg_quality": 95,
-      "enable_deskew": true,
-      "enable_contrast_enhance": false,
-      "enable_moire_removal": false
-    },
-    "lightweight": {
-      "description": "轻量级模式：适合质量已经较好的文档，仅做基础优化（数学公式密集文档推荐）",
-      "target_dpi": 150,
-      "max_side_len": 1920,
-      "jpeg_quality": 95,
-      "enable_deskew": true,
-      "enable_contrast_enhance": false,
-      "enable_moire_removal": false
-    },
-    "aggressive_compression": {
-      "description": "极限压缩模式：网络极差或对时间要求极高时的备选方案",
-      "target_dpi": 72,
-      "max_side_len": 1600,
-      "jpeg_quality": 85,
-      "enable_deskew": false,
-      "enable_contrast_enhance": false,
-      "enable_moire_removal": false
-    },
-    "illumination_rescue": {
-      "description": "光照拯救模式：针对阴影、反光、过曝等光照不均场景特调",
-      "target_dpi": 150,
-      "max_side_len": 1920,
-      "jpeg_quality": 92,
-      "enable_deskew": true,
-      "enable_contrast_enhance": true,
-      "enable_moire_removal": false,
-      "clahe_clip_limit": 1.5,
-      "clahe_tile_size": 8
-    },
-    "screen_photography": {
-      "description": "屏幕翻拍模式：专门处理电脑屏幕、投影仪翻拍产生的摩尔纹",
-      "target_dpi": 150,
-      "max_side_len": 2048,
-      "jpeg_quality": 95,
-      "enable_deskew": true,
-      "enable_contrast_enhance": false,
-      "enable_moire_removal": true
-    }
-  }
-}
\ No newline at end of file
diff --git a/data-processing/processing/evaluation_recognition.py b/data-processing/processing/evaluation_recognition.py
deleted file mode 100644
index f8dcecca..00000000
--- a/data-processing/processing/evaluation_recognition.py
+++ /dev/null
@@ -1,191 +0,0 @@
-import os
-import json
-import base64
-import requests
-import time
-import glob
-from concurrent.futures import ThreadPoolExecutor, as_completed
-import difflib
-from dotenv import load_dotenv
-
-load_dotenv()
-
-# ==========================================
-# 全局配置参数
-# ==========================================
-
-API_URL = os.getenv("PADDLEOCR_API_URL")
-TOKEN = os.getenv("PADDLEOCR_API_TOKEN")
-
-# 文件夹与文件配置
-GT_JSON_FILE = "OmniDocBench.json"                  # 官方下载的标注文件
-RAW_DIR = "sample_image/jiaocai"          # 存放样例原图的目录
-OUTPUT_LOG_FILE = "jiaocai_accuracy_results.json"       # 样例原图识别结果
-
-# 并发线程数 (不建议过高，避免触发 API 限流)
-MAX_WORKERS = 5
-
-# ==========================================
-# 模块一：解析 OmniDocBench 官方真值 (Ground Truth)
-# ==========================================
-def load_omnidocbench_gt(json_path):
-    print(f"正在加载官方标准答案库: {json_path}...")
-    if not os.path.exists(json_path):
-        raise FileNotFoundError(f"找不到文件 {json_path}，请确保已下载并放置在同级目录。")
-        
-    with open(json_path, 'r', encoding='utf-8') as f:
-        data = json.load(f)
-        
-    gt_dict = {}
-    for page in data:
-        img_name = os.path.basename(page['page_info']['image_path'])
-        
-        page_texts = []
-        for det in page.get('layout_dets', []):
-            if det.get('ignore', False): 
-                continue 
-                
-            if 'text' in det and det['text'].strip():
-                page_texts.append(det['text'].strip())
-            elif 'latex' in det and det['latex'].strip():
-                page_texts.append(det['latex'].strip())
-                
-        gt_dict[img_name] = "\n".join(page_texts)
-        
-    print(f"成功加载 {len(gt_dict)} 张图片的标准答案！\n")
-    return gt_dict
-
-def calculate_accuracy(gt_text, pred_text):
-    gt_clean = gt_text.strip().lower()
-    pred_clean = pred_text.strip().lower()
-    
-    if not gt_clean and not pred_clean: return 100.0
-    if not gt_clean or not pred_clean: return 0.0
-    
-    matcher = difflib.SequenceMatcher(None, gt_clean, pred_clean)
-    return round(matcher.ratio() * 100, 2)
-
-# ==========================================
-# 模块二：云端 API 调用
-# ==========================================
-def call_api(file_path):
-    with open(file_path, "rb") as f:
-        file_data = base64.b64encode(f.read()).decode("ascii")
-        
-    payload = {
-        "file": file_data, 
-        "fileType": 1, 
-        "useDocOrientationClassify": False, 
-        "useDocUnwarping": False
-    }
-    headers = {"Authorization": f"token {TOKEN}", "Content-Type": "application/json"}
-    
-    for i in range(3): 
-        try:
-            start = time.perf_counter()
-            response = requests.post(API_URL, json=payload, headers=headers, timeout=60)
-            response.raise_for_status()
-            latency = time.perf_counter() - start
-            return latency, response.json()
-        except Exception as e:
-            time.sleep(2 ** i)
-    return None, {"error": "API 请求失败"}
-
-# ==========================================
-# 模块三：多线程任务单元
-# ==========================================
-def process_single_task(img_path, gt_dict):
-    filename = os.path.basename(img_path)
-    file_size_kb = round(os.path.getsize(img_path) / 1024, 2)
-    
-    # Raw 文件名就是标准答案的 Key
-    if filename not in gt_dict:
-        return filename, {"error": f"未在官方 JSON 中找到对应的标准答案: {filename}"}
-        
-    gt_text = gt_dict[filename]
-    strategy = "raw"
-    
-    latency, res = call_api(img_path)
-    if not latency:
-        return filename, {"error": "API 调用失败"}
-        
-    try:
-        data = res.get('result', {})
-        layout = data.get('layoutParsingResults', [{}])[0]
-        pred_text = layout.get('markdown', {}).get('text', '')
-        blocks_count = len(layout.get('prunedResult', {}).get('parsing_res_list', []))
-    except Exception:
-        pred_text = ""
-        blocks_count = 0
-        
-    accuracy = calculate_accuracy(gt_text, pred_text)
-    
-    return filename, {
-        "original_img": filename,
-        "strategy": strategy,
-        "latency_s": round(latency, 2),
-        "size_kb": file_size_kb,
-        "text_len": len(pred_text),
-        "true_accuracy_pct": accuracy,
-        "blocks": blocks_count
-    }
-
-# ==========================================
-# 模块四：主控引擎 (多线程并发)
-# ==========================================
-def run_raw_evaluation():
-    try:
-        gt_dict = load_omnidocbench_gt(GT_JSON_FILE)
-    except Exception as e:
-        print(e)
-        return
-
-    final_results = {}
-    if os.path.exists(OUTPUT_LOG_FILE):
-        with open(OUTPUT_LOG_FILE, "r", encoding="utf-8") as f:
-            try:
-                final_results = json.load(f)
-                print(f"发现已有成绩单，已加载 {len(final_results)} 条记录。")
-            except:
-                pass
-
-    if not os.path.exists(RAW_DIR):
-        print(f"错误: 找不到文件夹 {RAW_DIR}，请检查路径。")
-        return
-        
-    files = glob.glob(os.path.join(RAW_DIR, "*.*"))
-    image_files = [f for f in files if f.lower().endswith(('.jpg', '.jpeg', '.png'))]
-            
-    # 剔除已完成
-    pending_files = [f for f in image_files if os.path.basename(f) not in final_results]
-    
-    if not pending_files:
-        print("所有 RAW 图片均已测试完毕！")
-        return
-    
-    total_files = len(pending_files)
-    print(f"开始测试 RAW 图片，共 {total_files} 个样本...")
-    
-    completed = 0
-    start_time = time.time()
-
-    with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
-        future_to_img = {executor.submit(process_single_task, img_path, gt_dict): img_path for img_path in pending_files}
-        
-        for future in as_completed(future_to_img):
-            filename, metrics = future.result()
-            final_results[filename] = metrics 
-            
-            completed += 1
-            acc = metrics.get('true_accuracy_pct', 'N/A')
-            lat = metrics.get('latency_s', 'N/A')
-            print(f"[{completed}/{total_files}] RAW 处理完成: {filename[:30]}... | 准确率: {acc}% | 耗时: {lat}s")
-            
-            with open(OUTPUT_LOG_FILE, "w", encoding="utf-8") as f:
-                json.dump(final_results, f, ensure_ascii=False, indent=4)
-
-    total_time = round(time.time() - start_time, 2)
-    print(f"\nRAW 图片测试全部完成！总耗时: {total_time} 秒")
-
-if __name__ == "__main__":
-    run_raw_evaluation()
\ No newline at end of file
diff --git a/data-processing/processing/exam_accuracy_results.json b/data-processing/processing/exam_accuracy_results.json
deleted file mode 100644
index ea1b1d64..00000000
--- a/data-processing/processing/exam_accuracy_results.json
+++ /dev/null
@@ -1,92 +0,0 @@
-{
-    "exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png",
-        "strategy": "raw",
-        "latency_s": 1.48,
-        "size_kb": 430.64,
-        "text_len": 3214,
-        "true_accuracy_pct": 78.42,
-        "blocks": 16
-    },
-    "exam_paper_2018年广西北海、钦州、南宁、来宾、崇左、防城港、北部湾经济区中考英语试题（空白卷）_page_001.png": {
-        "original_img": "exam_paper_2018年广西北海、钦州、南宁、来宾、崇左、防城港、北部湾经济区中考英语试题（空白卷）_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.62,
-        "size_kb": 448.54,
-        "text_len": 1178,
-        "true_accuracy_pct": 85.9,
-        "blocks": 16
-    },
-    "exam_paper_2016年安徽省中考英语试题（解析）_page_003.png": {
-        "original_img": "exam_paper_2016年安徽省中考英语试题（解析）_page_003.png",
-        "strategy": "raw",
-        "latency_s": 1.95,
-        "size_kb": 346.44,
-        "text_len": 899,
-        "true_accuracy_pct": 91.58,
-        "blocks": 13
-    },
-    "exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png",
-        "strategy": "raw",
-        "latency_s": 2.24,
-        "size_kb": 478.25,
-        "text_len": 3861,
-        "true_accuracy_pct": 78.96,
-        "blocks": 24
-    },
-    "exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.48,
-        "size_kb": 424.95,
-        "text_len": 3441,
-        "true_accuracy_pct": 79.24,
-        "blocks": 20
-    },
-    "exam_paper_2016年安徽省中考英语试题（解析）_page_015.png": {
-        "original_img": "exam_paper_2016年安徽省中考英语试题（解析）_page_015.png",
-        "strategy": "raw",
-        "latency_s": 3.12,
-        "size_kb": 417.37,
-        "text_len": 2060,
-        "true_accuracy_pct": 96.34,
-        "blocks": 12
-    },
-    "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png",
-        "strategy": "raw",
-        "latency_s": 1.63,
-        "size_kb": 642.68,
-        "text_len": 5019,
-        "true_accuracy_pct": 77.01,
-        "blocks": 29
-    },
-    "exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.63,
-        "size_kb": 479.37,
-        "text_len": 3534,
-        "true_accuracy_pct": 84.34,
-        "blocks": 29
-    },
-    "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png",
-        "strategy": "raw",
-        "latency_s": 1.98,
-        "size_kb": 714.66,
-        "text_len": 5844,
-        "true_accuracy_pct": 85.32,
-        "blocks": 24
-    },
-    "exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.35,
-        "size_kb": 426.23,
-        "text_len": 3329,
-        "true_accuracy_pct": 82.49,
-        "blocks": 27
-    }
-}
\ No newline at end of file
diff --git a/data-processing/processing/jiaocai_accuracy_results.json b/data-processing/processing/jiaocai_accuracy_results.json
deleted file mode 100644
index a4c0807f..00000000
--- a/data-processing/processing/jiaocai_accuracy_results.json
+++ /dev/null
@@ -1,92 +0,0 @@
-{
-    "jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg": {
-        "original_img": "jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg",
-        "strategy": "raw",
-        "latency_s": 1.1,
-        "size_kb": 182.96,
-        "text_len": 1440,
-        "true_accuracy_pct": 91.17,
-        "blocks": 9
-    },
-    "jiaocaineedrop_eng-45646.pdf_30.jpg": {
-        "original_img": "jiaocaineedrop_eng-45646.pdf_30.jpg",
-        "strategy": "raw",
-        "latency_s": 1.18,
-        "size_kb": 184.09,
-        "text_len": 2348,
-        "true_accuracy_pct": 93.59,
-        "blocks": 21
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1249.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1249.jpg",
-        "strategy": "raw",
-        "latency_s": 1.24,
-        "size_kb": 234.65,
-        "text_len": 2859,
-        "true_accuracy_pct": 95.9,
-        "blocks": 10
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1303.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1303.jpg",
-        "strategy": "raw",
-        "latency_s": 1.28,
-        "size_kb": 286.56,
-        "text_len": 1264,
-        "true_accuracy_pct": 91.93,
-        "blocks": 13
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1544.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1544.jpg",
-        "strategy": "raw",
-        "latency_s": 1.29,
-        "size_kb": 251.11,
-        "text_len": 1191,
-        "true_accuracy_pct": 91.86,
-        "blocks": 31
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1884.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1884.jpg",
-        "strategy": "raw",
-        "latency_s": 0.89,
-        "size_kb": 226.79,
-        "text_len": 980,
-        "true_accuracy_pct": 95.21,
-        "blocks": 38
-    },
-    "jiaocai_jiaocai_en_117.jpg": {
-        "original_img": "jiaocai_jiaocai_en_117.jpg",
-        "strategy": "raw",
-        "latency_s": 0.7,
-        "size_kb": 62.76,
-        "text_len": 188,
-        "true_accuracy_pct": 94.51,
-        "blocks": 7
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_2002.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_2002.jpg",
-        "strategy": "raw",
-        "latency_s": 0.97,
-        "size_kb": 259.87,
-        "text_len": 916,
-        "true_accuracy_pct": 96.73,
-        "blocks": 12
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_3428.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_3428.jpg",
-        "strategy": "raw",
-        "latency_s": 1.11,
-        "size_kb": 364.84,
-        "text_len": 1422,
-        "true_accuracy_pct": 96.99,
-        "blocks": 13
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_377.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_377.jpg",
-        "strategy": "raw",
-        "latency_s": 1.44,
-        "size_kb": 372.6,
-        "text_len": 5329,
-        "true_accuracy_pct": 96.81,
-        "blocks": 78
-    }
-}
\ No newline at end of file
diff --git a/data-processing/processing/sample_image/exam/exam_accuracy_results.json b/data-processing/processing/sample_image/exam/exam_accuracy_results.json
deleted file mode 100644
index fdadd52a..00000000
--- a/data-processing/processing/sample_image/exam/exam_accuracy_results.json
+++ /dev/null
@@ -1,92 +0,0 @@
-{
-    "exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png",
-        "strategy": "raw",
-        "latency_s": 1.95,
-        "size_kb": 430.64,
-        "text_len": 3214,
-        "true_accuracy_pct": 78.42,
-        "blocks": 16
-    },
-    "exam_paper_2016年安徽省中考英语试题（解析）_page_015.png": {
-        "original_img": "exam_paper_2016年安徽省中考英语试题（解析）_page_015.png",
-        "strategy": "raw",
-        "latency_s": 1.97,
-        "size_kb": 417.37,
-        "text_len": 2060,
-        "true_accuracy_pct": 96.34,
-        "blocks": 12
-    },
-    "exam_paper_2016年安徽省中考英语试题（解析）_page_003.png": {
-        "original_img": "exam_paper_2016年安徽省中考英语试题（解析）_page_003.png",
-        "strategy": "raw",
-        "latency_s": 2.02,
-        "size_kb": 346.44,
-        "text_len": 899,
-        "true_accuracy_pct": 91.58,
-        "blocks": 13
-    },
-    "exam_paper_2018年广西北海、钦州、南宁、来宾、崇左、防城港、北部湾经济区中考英语试题（空白卷）_page_001.png": {
-        "original_img": "exam_paper_2018年广西北海、钦州、南宁、来宾、崇左、防城港、北部湾经济区中考英语试题（空白卷）_page_001.png",
-        "strategy": "raw",
-        "latency_s": 2.23,
-        "size_kb": 448.54,
-        "text_len": 1178,
-        "true_accuracy_pct": 85.9,
-        "blocks": 16
-    },
-    "exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png",
-        "strategy": "raw",
-        "latency_s": 2.26,
-        "size_kb": 478.25,
-        "text_len": 3861,
-        "true_accuracy_pct": 78.96,
-        "blocks": 24
-    },
-    "exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.75,
-        "size_kb": 424.95,
-        "text_len": 3441,
-        "true_accuracy_pct": 79.24,
-        "blocks": 20
-    },
-    "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png",
-        "strategy": "raw",
-        "latency_s": 1.98,
-        "size_kb": 642.68,
-        "text_len": 5019,
-        "true_accuracy_pct": 77.01,
-        "blocks": 29
-    },
-    "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png",
-        "strategy": "raw",
-        "latency_s": 2.09,
-        "size_kb": 714.66,
-        "text_len": 5844,
-        "true_accuracy_pct": 85.32,
-        "blocks": 24
-    },
-    "exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png",
-        "strategy": "raw",
-        "latency_s": 1.95,
-        "size_kb": 479.37,
-        "text_len": 3534,
-        "true_accuracy_pct": 84.34,
-        "blocks": 29
-    },
-    "exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png": {
-        "original_img": "exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png",
-        "strategy": "raw",
-        "latency_s": 2.04,
-        "size_kb": 426.23,
-        "text_len": 3329,
-        "true_accuracy_pct": 82.49,
-        "blocks": 27
-    }
-}
\ No newline at end of file
diff --git "a/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_003.png" "b/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_003.png"
deleted file mode 100644
index f6097dc7..00000000
Binary files "a/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_003.png" and /dev/null differ
diff --git "a/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_015.png" "b/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_015.png"
deleted file mode 100644
index c6beb4ef..00000000
Binary files "a/data-processing/processing/sample_image/exam/exam_paper_2016\345\271\264\345\256\211\345\276\275\347\234\201\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\350\247\243\346\236\220\357\274\211_page_015.png" and /dev/null differ
diff --git "a/data-processing/processing/sample_image/exam/exam_paper_2018\345\271\264\345\271\277\350\245\277\345\214\227\346\265\267\343\200\201\351\222\246\345\267\236\343\200\201\345\215\227\345\256\201\343\200\201\346\235\245\345\256\276\343\200\201\345\264\207\345\267\246\343\200\201\351\230\262\345\237\216\346\270\257\343\200\201\345\214\227\351\203\250\346\271\276\347\273\217\346\265\216\345\214\272\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\347\251\272\347\231\275\345\215\267\357\274\211_page_001.png" "b/data-processing/processing/sample_image/exam/exam_paper_2018\345\271\264\345\271\277\350\245\277\345\214\227\346\265\267\343\200\201\351\222\246\345\267\236\343\200\201\345\215\227\345\256\201\343\200\201\346\235\245\345\256\276\343\200\201\345\264\207\345\267\246\343\200\201\351\230\262\345\237\216\346\270\257\343\200\201\345\214\227\351\203\250\346\271\276\347\273\217\346\265\216\345\214\272\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\347\251\272\347\231\275\345\215\267\357\274\211_page_001.png"
deleted file mode 100644
index 97183707..00000000
Binary files "a/data-processing/processing/sample_image/exam/exam_paper_2018\345\271\264\345\271\277\350\245\277\345\214\227\346\265\267\343\200\201\351\222\246\345\267\236\343\200\201\345\215\227\345\256\201\343\200\201\346\235\245\345\256\276\343\200\201\345\264\207\345\267\246\343\200\201\351\230\262\345\237\216\346\270\257\343\200\201\345\214\227\351\203\250\346\271\276\347\273\217\346\265\216\345\214\272\344\270\255\350\200\203\350\213\261\350\257\255\350\257\225\351\242\230\357\274\210\347\251\272\347\231\275\345\215\267\357\274\211_page_001.png" and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png
deleted file mode 100644
index ac8aa32e..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_1997_Solutions_1997s_page_003.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png
deleted file mode 100644
index 27c24607..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2006_Problems_2006_page_001.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png
deleted file mode 100644
index 06967a96..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2007_Problems_2007_page_001.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png
deleted file mode 100644
index 93eac5c9..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_002.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png
deleted file mode 100644
index 6ea86371..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2009_Solutions_2009s_page_003.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png
deleted file mode 100644
index 16062a55..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2014_Problems_2014_page_001.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png b/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png
deleted file mode 100644
index 564ef041..00000000
Binary files a/data-processing/processing/sample_image/exam/exam_paper_en-file-putnam-archive_2015_Problems_2015_page_001.png and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocai_accuracy_results.json b/data-processing/processing/sample_image/jiaocai/jiaocai_accuracy_results.json
deleted file mode 100644
index 2cb21262..00000000
--- a/data-processing/processing/sample_image/jiaocai/jiaocai_accuracy_results.json
+++ /dev/null
@@ -1,92 +0,0 @@
-{
-    "jiaocaineedrop_jiaocai_needrop_en_1249.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1249.jpg",
-        "strategy": "raw",
-        "latency_s": 0.99,
-        "size_kb": 234.65,
-        "text_len": 2859,
-        "true_accuracy_pct": 95.9,
-        "blocks": 10
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1544.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1544.jpg",
-        "strategy": "raw",
-        "latency_s": 1.09,
-        "size_kb": 251.11,
-        "text_len": 1191,
-        "true_accuracy_pct": 91.86,
-        "blocks": 31
-    },
-    "jiaocaineedrop_eng-45646.pdf_30.jpg": {
-        "original_img": "jiaocaineedrop_eng-45646.pdf_30.jpg",
-        "strategy": "raw",
-        "latency_s": 1.11,
-        "size_kb": 184.09,
-        "text_len": 2348,
-        "true_accuracy_pct": 93.59,
-        "blocks": 21
-    },
-    "jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg": {
-        "original_img": "jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg",
-        "strategy": "raw",
-        "latency_s": 1.85,
-        "size_kb": 182.96,
-        "text_len": 1440,
-        "true_accuracy_pct": 91.17,
-        "blocks": 9
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1884.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1884.jpg",
-        "strategy": "raw",
-        "latency_s": 1.06,
-        "size_kb": 226.79,
-        "text_len": 980,
-        "true_accuracy_pct": 95.21,
-        "blocks": 38
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_1303.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_1303.jpg",
-        "strategy": "raw",
-        "latency_s": 2.11,
-        "size_kb": 286.56,
-        "text_len": 1264,
-        "true_accuracy_pct": 91.93,
-        "blocks": 13
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_2002.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_2002.jpg",
-        "strategy": "raw",
-        "latency_s": 1.18,
-        "size_kb": 259.87,
-        "text_len": 916,
-        "true_accuracy_pct": 96.73,
-        "blocks": 12
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_3428.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_3428.jpg",
-        "strategy": "raw",
-        "latency_s": 1.27,
-        "size_kb": 364.84,
-        "text_len": 1422,
-        "true_accuracy_pct": 96.99,
-        "blocks": 13
-    },
-    "jiaocaineedrop_jiaocai_needrop_en_377.jpg": {
-        "original_img": "jiaocaineedrop_jiaocai_needrop_en_377.jpg",
-        "strategy": "raw",
-        "latency_s": 1.46,
-        "size_kb": 372.6,
-        "text_len": 5329,
-        "true_accuracy_pct": 96.81,
-        "blocks": 78
-    },
-    "jiaocai_jiaocai_en_117.jpg": {
-        "original_img": "jiaocai_jiaocai_en_117.jpg",
-        "strategy": "raw",
-        "latency_s": 0.6,
-        "size_kb": 62.76,
-        "text_len": 188,
-        "true_accuracy_pct": 94.51,
-        "blocks": 7
-    }
-}
\ No newline at end of file
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocai_jiaocai_en_117.jpg b/data-processing/processing/sample_image/jiaocai/jiaocai_jiaocai_en_117.jpg
deleted file mode 100644
index 7419de25..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocai_jiaocai_en_117.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg
deleted file mode 100644
index a00788a4..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_624b60c58c9d8bfb6ff1886c2fd605d2adeb6ea4da576068201b6c6958ce93f4.pdf_8.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_eng-45646.pdf_30.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_eng-45646.pdf_30.jpg
deleted file mode 100644
index e4c93db0..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_eng-45646.pdf_30.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1249.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1249.jpg
deleted file mode 100644
index 3548fc85..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1249.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1303.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1303.jpg
deleted file mode 100644
index efdc7cda..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1303.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1544.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1544.jpg
deleted file mode 100644
index 703881c7..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1544.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1884.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1884.jpg
deleted file mode 100644
index b3d81f4c..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_1884.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_2002.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_2002.jpg
deleted file mode 100644
index 8c5251bd..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_2002.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_3428.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_3428.jpg
deleted file mode 100644
index 5ff6f3c2..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_3428.jpg and /dev/null differ
diff --git a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_377.jpg b/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_377.jpg
deleted file mode 100644
index 95d2faaa..00000000
Binary files a/data-processing/processing/sample_image/jiaocai/jiaocaineedrop_jiaocai_needrop_en_377.jpg and /dev/null differ
diff --git a/data-processing/standard_doc/DATA_LABELING_STANDARDS.md b/data-processing/standard_doc/DATA_LABELING_STANDARDS.md
deleted file mode 100644
index c9722099..00000000
--- a/data-processing/standard_doc/DATA_LABELING_STANDARDS.md
+++ /dev/null
@@ -1,62 +0,0 @@
-# 题目标注格式与数据规范 (V1.0)
-
-本规范旨在定义试卷经过 PaddleOCR 解析后，如何将非结构化的文本内容标准化为可供数据库存储、搜索引擎检索及 AI 模型使用的结构化数据（JSON 格式）。
-
-## 一、 核心字段定义规范
-
-无论什么题型（单选、多选、填空、解答），一道标准化的题目必须包含以下核心结构：
-
-1. **`question_type` (题型)**
-   * **枚举值**：`single_choice` (单选), `multiple_choice` (多选), `fill_blank` (填空), `essay` (解答题/主观题)。
-
-2. **`stem` (题干)**
-   * **规范**：包含题目主体描述。如果是填空题，填空处强制统一使用 `___`（三个下划线）占位。
-
-3. **`options` (选项 - 仅限选择题)**
-   * **规范**：必须是一个包含完整选项的数组（Array）或键值对（Object）。禁止将选项 A、B、C 杂糅在题干文本中。
-
-4. **`answer` (标准答案)**
-   * **规范**：
-     * 选择题：存大写字母，如 `"A"` 或 `["A", "C"]`。
-     * 填空题：按空序存储字符串数组，如 `["3.14", "牛顿"]`。
-
-5. **`analysis` (解析)**
-   * **规范**：详细的解题步骤。可选字段，若原卷未提供可置为空串。
-
-## 二、 富文本与多媒体格式规范
-
-试卷中往往包含大量的公式和配图，必须严格按照以下格式进行转录和标注：
-
-1. **数学与物理公式 (Math & Physics)**
-   * **行内公式**：强制使用单美元符号包裹。例：`设圆的半径为 $R$`。
-   * **独立段落公式**：强制使用双美元符号包裹。例：`$$E = mc^2$$`
-   * **禁忌**：严禁混用 Unicode 特殊字符（如使用 `㎡` 代替 `$m^2$`，使用 `α` 代替 `$\alpha$`），一切以 LaTeX 语法为准。
-
-2. **配图与占位 (Images)**
-   * **格式**：强制使用标准 Markdown 图片语法 `![图片描述](图片本地或OSS存储路径)`。
-   * **命名规范**：图片路径需与题目 ID 强绑定，例如 `![题1配图](/images/q_1001_fig1.jpg)`。
-
-## 三、 标准化 JSON 结构示例
-
-在完成人工校对或大模型结构化提取后，最终入库的数据必须符合以下 JSON Schema 结构：
-
-### 示例：一道标准的单选题
-
-```json
-{
-  "question_id": "Q_2023_MATH_001",
-  "question_type": "single_choice",
-  "difficulty": "medium",
-  "knowledge_tags": ["二次函数", "极值"],
-  "content": {
-    "stem": "已知二次函数 $y = ax^2 + bx + c$ 的图像如图所示，则下列结论正确的是？\n![二次函数图](/assets/images/q001.jpg)",
-    "options": {
-      "A": "$a > 0$",
-      "B": "$b < 0$",
-      "C": "$c = 0$",
-      "D": "$b^2 - 4ac < 0$"
-    }
-  },
-  "answer": "A",
-  "analysis": "由图像开口向上可知 $a > 0$，故选 A。"
-}
\ No newline at end of file
diff --git a/data-processing/standard_doc/SCANNING_STANDARDS.md b/data-processing/standard_doc/SCANNING_STANDARDS.md
deleted file mode 100644
index 118d4c79..00000000
--- a/data-processing/standard_doc/SCANNING_STANDARDS.md
+++ /dev/null
@@ -1,32 +0,0 @@
-# 试卷数字化扫描规范 (V1.0)
-
-基于对  [Real5-OmniDocBench](https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench)场景的本地初步实验验证，为最大化 PaddleOCR 引擎的识别精度并最小化网络与计算延迟，特制定本参数标准。
-
-## 一、 确定分辨率要求
-
-禁止扫描设备直接输出 4K (3840px) 原始大图。过大的分辨率会引入背景高频噪点并造成极大的传输浪费。
-
-* **推荐长边像素**：**1920px**
-* **极限上限像素**：**2048px**
-  * *说明：1080P 级别是 PaddleOCR 识别试卷结构和文本的最佳感受野，能拦截大量环境噪点，且传输耗时大幅降低70%。*
-* **物理像素密度**：**150 DPI**
-  * *说明：试卷包含大量密集小字与数学公式。72 DPI 易导致笔画粘连，300 DPI 则会徒增云端显存压力，150 DPI 为实测最优解。*
-
-## 二、 确定格式要求
-
-* **强制输出格式**：**JPEG**
-* **压缩质量 (Quality)**：**Q90 - Q95**
-  * *说明：在 1920px 尺度下，Q90 是兼顾“极小文件体积（省带宽）”与“零精度损耗（保识别）”的最佳平衡点。*
-
-## 三、 硬件采集防失真规范
-
-前端采集必须从物理层面规避以下四种典型的试卷失真类型：
-
-1. **光照不均 (Illumination)**：必须采用均匀的顶部或双侧布光，严禁试卷局部被阴影遮挡或桌面强烈反光。
-2. **变形褶皱 (Warping)**：必须确保试卷完全展平，严禁边缘卷曲或装订线中缝深陷。
-3. **倾斜 (Skew)**：试卷边缘与取景框应保持平行，物理倾斜角严格控制在 **±5°** 以内。
-4. **屏幕翻拍 (Screen-Photography)**：强制直接扫描物理纸张，严禁对着电脑屏幕或投影仪进行二次翻拍（会产生致命摩尔纹）。
-
-## 四、 其他
-
-若受客观环境限制（如考场恶劣光线下的随手拍），必须在调用 PaddleOCR API 前，接入自适应预处理诊断模块（`mode="auto"`），利用 CLAHE 光照增强和自动降维来挽救极端畸变样本。
\ No newline at end of file
diff --git a/docs/API.md b/docs/API.md
deleted file mode 100644
index c289e00e..00000000
--- a/docs/API.md
+++ /dev/null
@@ -1,571 +0,0 @@
-# API 接口文档
-
-错题本生成系统提供 RESTful API 接口，用于文件上传、OCR 解析、题目分割和错题本导出。
-
-**Base URL**: `http://localhost:5001`
-
----
-
-## 目录
-
-- [概览](#概览)
-- [接口列表](#接口列表)
-  - [GET / — 主页](#get-——主页)
-  - [GET /api/status — 系统状态](#get-apistatus——系统状态)
-  - [POST /api/upload — 上传文件](#post-apiupload——上传文件)
-  - [POST /api/split — 分割题目](#post-apisplit——分割题目)
-  - [POST /api/export — 导出错题本](#post-apiexport——导出错题本)
-  - [GET /api/questions — 获取题目列表](#get-apiquestions——获取题目列表)
-  - [GET /preview — 预览页面](#get-preview——预览页面)
-  - [GET /download/\<filename\> — 下载文件](#get-downloadfilename——下载文件)
-- [数据结构](#数据结构)
-- [调用流程](#调用流程)
-- [错误处理](#错误处理)
-
----
-
-## 概览
-
-### 技术栈
-
-- **框架**: Flask 3.1.2
-- **状态管理**: LangGraph MemorySaver（基于 `thread_id` 的会话状态）
-- **工作流**: LangGraph StateGraph，在 `split_questions` 和 `export` 节点前设置中断
-
-### 典型调用流程
-
-```
-1. GET  /api/status          ← 检查系统配置
-2. POST /api/upload           ← 上传文件，触发 prepare_input + ocr_parse
-3. POST /api/split            ← 恢复工作流，执行 Agent 题目分割
-4. POST /api/export           ← 注入选中题目 ID，执行导出
-5. GET  /download/wrongbook.md ← 下载导出的错题本
-```
-
-### 通用响应格式
-
-成功响应：
-```json
-{
-  "success": true,
-  "message": "操作描述",
-  ...
-}
-```
-
-错误响应：
-```json
-{
-  "success": false,
-  "error": "错误描述"
-}
-```
-
----
-
-## 接口列表
-
-### GET / — 主页
-
-返回 Web 界面主页。
-
-**请求**:
-```
-GET /
-```
-
-**响应**: HTML 页面（`templates/index.html`）
-
----
-
-### GET /api/status — 系统状态
-
-获取系统配置状态，检查各项外部服务是否已正确配置。
-
-**请求**:
-```
-GET /api/status
-```
-
-**响应示例**:
-```json
-{
-  "success": true,
-  "status": {
-    "paddleocr_configured": true,
-    "deepseek_configured": true,
-    "langsmith_enabled": false,
-    "output_dirs": {
-      "pages": "output/pages",
-      "struct": "output/struct",
-      "results": "results"
-    }
-  }
-}
-```
-
-**响应字段说明**:
-
-| 字段 | 类型 | 说明 |
-|------|------|------|
-| `status.paddleocr_configured` | boolean | PaddleOCR API 是否已配置（检查 `PADDLEOCR_API_URL` 环境变量） |
-| `status.deepseek_configured` | boolean | DeepSeek API 是否已配置（检查 `DEEPSEEK_API_KEY` 环境变量） |
-| `status.langsmith_enabled` | boolean | LangSmith 追踪是否已启用 |
-| `status.output_dirs` | object | 输出目录路径配置 |
-
----
-
-### POST /api/upload — 上传文件
-
-上传 PDF 或图片文件，系统自动执行预处理流程：
-
-1. 保存上传文件到 `uploads/` 目录
-2. 创建新的工作流会话（`thread_id`）
-3. 执行 `prepare_input` 节点 — 将文件转换为标准化 PNG 图片
-4. 执行 `ocr_parse` 节点 — 调用 PaddleOCR API 解析文档结构
-5. 在 `split_questions` 节点前中断，等待用户触发
-
-**请求**:
-```
-POST /api/upload
-Content-Type: multipart/form-data
-```
-
-**请求参数**:
-
-| 参数 | 类型 | 必需 | 说明 |
-|------|------|------|------|
-| `file` | File | 是 | 上传的文件（PDF 或图片） |
-
-**支持的文件格式**: `.pdf`, `.png`, `.jpg`, `.jpeg`, `.bmp`, `.tiff`, `.webp`
-
-**文件大小限制**: 50MB
-
-**成功响应** (200):
-```json
-{
-  "success": true,
-  "message": "文件处理成功",
-  "result": {
-    "image_count": 3,
-    "ocr_count": 3,
-    "image_paths": [
-      "output/pages/test_page_001.png",
-      "output/pages/test_page_002.png",
-      "output/pages/test_page_003.png"
-    ]
-  }
-}
-```
-
-**响应字段说明**:
-
-| 字段 | 类型 | 说明 |
-|------|------|------|
-| `result.image_count` | number | 转换后的图片数量 |
-| `result.ocr_count` | number | OCR 解析完成的图片数量 |
-| `result.image_paths` | array | 标准化后的图片路径列表 |
-
-**错误响应**:
-
-| 状态码 | 错误 | 说明 |
-|--------|------|------|
-| 400 | `没有上传文件` | 请求中未包含文件 |
-| 400 | `未选择文件` | 文件名为空 |
-| 400 | `不支持的文件格式` | 文件扩展名不在允许列表中 |
-| 413 | — | 文件超过 50MB 大小限制（Flask 自动返回） |
-| 500 | `{错误详情}` | 服务端处理异常 |
-
-**cURL 示例**:
-```bash
-curl -X POST http://localhost:5001/api/upload \
-  -F "file=@/path/to/exam.pdf"
-```
-
----
-
-### POST /api/split — 分割题目
-
-恢复工作流执行 Agent 题目分割。必须在 `/api/upload` 成功后调用。
-
-**内部流程**:
-1. 使用当前会话的 `thread_id` 恢复工作流
-2. 执行 `split_questions` 节点 — DeepSeek Agent 分析 OCR 结果并分割题目
-3. 在 `export` 节点前中断，等待用户选择题目
-
-**请求**:
-```
-POST /api/split
-```
-
-无需请求体。
-
-**成功响应** (200):
-```json
-{
-  "success": true,
-  "message": "成功分割 5 道题目",
-  "questions": [
-    {
-      "question_id": "1",
-      "question_type": "选择题",
-      "content_blocks": [
-        {
-          "block_type": "text",
-          "content": "下列关于函数的说法，正确的是（  ）",
-          "bbox": [100, 200, 800, 250],
-          "block_id": 1
-        }
-      ],
-      "options": [
-        "A. 函数的图像关于y轴对称",
-        "B. 函数是增函数",
-        "C. 函数的最小值为1",
-        "D. 函数的周期为π"
-      ],
-      "has_formula": false,
-      "has_image": false,
-      "image_refs": []
-    }
-  ]
-}
-```
-
-**错误响应**:
-
-| 状态码 | 错误 | 说明 |
-|--------|------|------|
-| 400 | `请先上传文件` | 未先调用 `/api/upload` |
-| 500 | `{错误详情}` | Agent 执行异常 |
-
-**cURL 示例**:
-```bash
-curl -X POST http://localhost:5001/api/split
-```
-
----
-
-### POST /api/export — 导出错题本
-
-注入用户选中的题目 ID，恢复工作流执行导出，生成 Markdown 格式的错题本。
-
-**内部流程**:
-1. 将 `selected_ids` 注入工作流状态
-2. 恢复工作流执行 `export` 节点
-3. 生成 `results/wrongbook.md` 文件
-
-**请求**:
-```
-POST /api/export
-Content-Type: application/json
-```
-
-**请求体**:
-```json
-{
-  "selected_ids": ["1", "3", "5"]
-}
-```
-
-**请求参数**:
-
-| 字段 | 类型 | 必需 | 说明 |
-|------|------|------|------|
-| `selected_ids` | array of string | 是 | 要导出的题目 ID 列表 |
-
-**成功响应** (200):
-```json
-{
-  "success": true,
-  "message": "错题本导出成功",
-  "output_path": "results/wrongbook.md"
-}
-```
-
-**错误响应**:
-
-| 状态码 | 错误 | 说明 |
-|--------|------|------|
-| 400 | `未选择任何题目` | `selected_ids` 为空 |
-| 400 | `请先分割题目` | 未先完成上传和分割流程 |
-| 500 | `{错误详情}` | 导出处理异常 |
-
-**cURL 示例**:
-```bash
-curl -X POST http://localhost:5001/api/export \
-  -H "Content-Type: application/json" \
-  -d '{"selected_ids": ["1", "2", "3"]}'
-```
-
----
-
-### GET /api/questions — 获取题目列表
-
-从 `results/questions.json` 文件读取已分割的题目列表。可在 `/api/split` 成功后随时调用获取题目数据。
-
-**请求**:
-```
-GET /api/questions
-```
-
-**成功响应** (200):
-```json
-{
-  "success": true,
-  "questions": [
-    {
-      "question_id": "1",
-      "question_type": "选择题",
-      "content_blocks": [...],
-      "options": [...],
-      "has_formula": true,
-      "has_image": false,
-      "image_refs": []
-    }
-  ]
-}
-```
-
-**无数据时的响应**:
-```json
-{
-  "success": true,
-  "questions": [],
-  "message": "暂无题目数据"
-}
-```
-
-**cURL 示例**:
-```bash
-curl http://localhost:5001/api/questions
-```
-
----
-
-### GET /preview — 预览页面
-
-返回 `results/preview.html` 的内容。如果预览文件不存在，返回 404。
-
-**请求**:
-```
-GET /preview
-```
-
-**响应**: HTML 页面内容
-
-**错误响应**:
-
-| 状态码 | 说明 |
-|--------|------|
-| 404 | `预览文件不存在，请先分割题目` |
-
----
-
-### GET /download/\<filename\> — 下载文件
-
-从 `results/` 目录下载指定文件。
-
-**请求**:
-```
-GET /download/<filename>
-```
-
-**路径参数**:
-
-| 参数 | 说明 |
-|------|------|
-| `filename` | 要下载的文件名（如 `wrongbook.md`、`questions.json`） |
-
-**响应**: 文件下载（`Content-Disposition: attachment`）
-
-**常用下载路径**:
-- `/download/wrongbook.md` — 错题本 Markdown 文件
-- `/download/questions.json` — 题目 JSON 数据
-- `/download/split_issues.jsonl` — 处理问题日志
-
-**cURL 示例**:
-```bash
-curl -O http://localhost:5001/download/wrongbook.md
-```
-
----
-
-## 数据结构
-
-### Question 对象
-
-Agent 分割后返回的题目数据结构：
-
-```json
-{
-  "question_id": "1",
-  "question_type": "选择题",
-  "content_blocks": [
-    {
-      "block_type": "text",
-      "content": "题干内容...",
-      "bbox": [x1, y1, x2, y2],
-      "block_id": 1
-    },
-    {
-      "block_type": "display_formula",
-      "content": "x^2 + y^2 = r^2",
-      "bbox": [x1, y1, x2, y2],
-      "block_id": 2
-    },
-    {
-      "block_type": "image",
-      "content": "output/assets/figure_1.jpg",
-      "bbox": [x1, y1, x2, y2],
-      "block_id": 3
-    }
-  ],
-  "options": ["A. 选项1", "B. 选项2", "C. 选项3", "D. 选项4"],
-  "has_formula": true,
-  "has_image": true,
-  "image_refs": ["output/assets/figure_1.jpg"]
-}
-```
-
-**字段说明**:
-
-| 字段 | 类型 | 说明 |
-|------|------|------|
-| `question_id` | string | 题号标识 |
-| `question_type` | string | 题型：`选择题` / `填空题` / `解答题` / `判断题` |
-| `content_blocks` | array | 题目内容块列表，按 `block_id` 排序 |
-| `options` | array | 选项列表（仅选择题有值） |
-| `has_formula` | boolean | 是否包含数学公式 |
-| `has_image` | boolean | 是否包含图片 |
-| `image_refs` | array | 图片引用路径列表 |
-
-### ContentBlock 对象
-
-题目内容块的数据结构：
-
-| 字段 | 类型 | 说明 |
-|------|------|------|
-| `block_type` | string | 块类型：`text` / `display_formula` / `inline_formula` / `image` |
-| `content` | string | 块内容（文字、LaTeX 公式或图片路径） |
-| `bbox` | array | 边界框坐标 `[x1, y1, x2, y2]` |
-| `block_id` | number | 块在题目中的顺序编号 |
-
-**block_type 取值说明**:
-
-| 值 | 说明 | content 格式 |
-|----|------|--------------|
-| `text` | 纯文字内容 | 普通文本 |
-| `display_formula` | 行间公式 | LaTeX 公式（不含 `$$`） |
-| `inline_formula` | 行内公式 | LaTeX 公式（不含 `$`） |
-| `image` | 图片引用 | 图片文件路径 |
-
-### WorkflowState
-
-LangGraph 工作流内部状态定义（`src/workflow.py`）：
-
-| 字段 | 类型 | 说明 |
-|------|------|------|
-| `file_path` | string | 上传的原始文件路径 |
-| `image_paths` | array | 标准化后的图片路径列表 |
-| `ocr_results` | array | PaddleOCR 解析结果列表 |
-| `questions` | array | Agent 分割后的题目列表 |
-| `selected_ids` | array | 用户选中的题目 ID |
-| `output_path` | string | 导出文件的路径 |
-
----
-
-## 调用流程
-
-### 完整流程示例（Python）
-
-```python
-import requests
-
-BASE_URL = "http://localhost:5001"
-
-# 1. 检查系统状态
-status = requests.get(f"{BASE_URL}/api/status").json()
-print("系统状态:", status["status"])
-
-# 2. 上传文件
-with open("exam.pdf", "rb") as f:
-    resp = requests.post(f"{BASE_URL}/api/upload", files={"file": f})
-upload_result = resp.json()
-print(f"识别到 {upload_result['result']['image_count']} 页")
-
-# 3. 分割题目
-resp = requests.post(f"{BASE_URL}/api/split")
-split_result = resp.json()
-questions = split_result["questions"]
-print(f"分割出 {len(questions)} 道题目")
-
-# 4. 选择并导出
-selected = [q["question_id"] for q in questions]  # 全选
-resp = requests.post(
-    f"{BASE_URL}/api/export",
-    json={"selected_ids": selected}
-)
-print("导出结果:", resp.json()["message"])
-
-# 5. 下载错题本
-resp = requests.get(f"{BASE_URL}/download/wrongbook.md")
-with open("wrongbook.md", "wb") as f:
-    f.write(resp.content)
-print("错题本已下载")
-```
-
-### 完整流程示例（cURL）
-
-```bash
-# 1. 检查系统状态
-curl http://localhost:5001/api/status
-
-# 2. 上传文件
-curl -X POST http://localhost:5001/api/upload \
-  -F "file=@exam.pdf"
-
-# 3. 分割题目
-curl -X POST http://localhost:5001/api/split
-
-# 4. 导出错题本
-curl -X POST http://localhost:5001/api/export \
-  -H "Content-Type: application/json" \
-  -d '{"selected_ids": ["1", "2", "3"]}'
-
-# 5. 下载错题本
-curl -O http://localhost:5001/download/wrongbook.md
-```
-
----
-
-## 错误处理
-
-### HTTP 状态码
-
-| 状态码 | 说明 |
-|--------|------|
-| 200 | 请求成功 |
-| 400 | 客户端错误（参数缺失、文件格式不支持等） |
-| 404 | 资源不存在（预览文件未生成等） |
-| 413 | 上传文件超过大小限制 |
-| 500 | 服务端内部错误 |
-
-### 常见错误场景
-
-| 场景 | 错误信息 | 解决方法 |
-|------|----------|----------|
-| 未上传文件直接分割 | `请先上传文件` | 先调用 `/api/upload` |
-| 未分割直接导出 | `请先分割题目` | 先调用 `/api/split` |
-| 上传不支持的格式 | `不支持的文件格式` | 使用支持的格式上传 |
-| 导出时未选题目 | `未选择任何题目` | `selected_ids` 不能为空 |
-| API 密钥未配置 | 500 错误 | 检查 `.env` 中的 API 密钥 |
-
-### 注意事项
-
-1. **会话状态**: 系统使用全局 `thread_id` 管理会话，一次只能处理一个文件。新的上传会覆盖之前的会话状态。
-2. **调用顺序**: 必须按 `upload → split → export` 的顺序调用，不能跳步。
-3. **并发限制**: 当前实现使用全局变量管理会话，不支持多用户并发访问。生产环境部署需要改造为基于请求的会话管理。
-
----
-
-*文档版本: v1.0.0*
diff --git a/docs/HANDWRITTEN_API.md b/docs/HANDWRITTEN_API.md
deleted file mode 100644
index 83768047..00000000
--- a/docs/HANDWRITTEN_API.md
+++ /dev/null
@@ -1,287 +0,0 @@
-# 手写体 vs 印刷体识别 API
-
-## 概述
-
-`printed-vs-handwritten/app.py` 提供了一个 Flask Web API，用于分类图片内容是手写还是印刷体。
-
-## 技术架构
-
-- **框架**: Flask
-- **模型**: TypegroupsClassifier (OCR-D typegroups classifier)
-- **端口**: 5000
-- **模型路径**: `ocrd_typegroups_classifier/models/classifier.tgc`
-
-## API 端点
-
-### 1. 健康检查
-
-```http
-GET /
-```
-
-**响应示例**:
-```json
-{
-  "status": "ok",
-  "message": "Printed vs Handwritten Classification API",
-  "endpoints": {
-    "/classify": "POST - Classify an image (multipart/form-data with 'image' field)",
-    "/classify_base64": "POST - Classify a base64 encoded image (JSON with 'image' field)"
-  }
-}
-```
-
-### 2. 分类图片（文件上传）
-
-```http
-POST /classify
-Content-Type: multipart/form-data
-```
-
-**请求参数**:
-- `image`: 图片文件 (multipart/form-data)
-
-**响应示例**:
-```json
-{
-  "success": true,
-  "prediction": "printed",
-  "confidence": 0.9523,
-  "probabilities": {
-    "printed": 0.9523,
-    "handwritten": 0.0477
-  }
-}
-```
-
-### 3. 分类图片（Base64）
-
-```http
-POST /classify_base64
-Content-Type: application/json
-```
-
-**请求体**:
-```json
-{
-  "image": "base64_encoded_image_string"
-}
-```
-
-**响应示例**:
-```json
-{
-  "success": true,
-  "prediction": "handwritten",
-  "confidence": 0.8745,
-  "probabilities": {
-    "printed": 0.1255,
-    "handwritten": 0.8745
-  }
-}
-```
-
-## 分类结果
-
-### 可能的预测值
-
-- `"printed"` - 印刷体
-- `"handwritten"` - 手写体
-
-### 置信度
-
-- `confidence`: 预测类别的置信度 (0-1)
-- `probabilities`: 所有类别的概率分布
-
-## 使用示例
-
-### Python 调用示例
-
-#### 方式1: 文件上传
-
-```python
-import requests
-
-url = "http://localhost:5000/classify"
-files = {"image": open("test_image.jpg", "rb")}
-
-response = requests.post(url, files=files)
-result = response.json()
-
-print(f"预测结果: {result['prediction']}")
-print(f"置信度: {result['confidence']:.2%}")
-```
-
-#### 方式2: Base64 编码
-
-```python
-import requests
-import base64
-
-# 读取并编码图片
-with open("test_image.jpg", "rb") as f:
-    image_data = base64.b64encode(f.read()).decode("ascii")
-
-url = "http://localhost:5000/classify_base64"
-payload = {"image": image_data}
-
-response = requests.post(url, json=payload)
-result = response.json()
-
-print(f"预测结果: {result['prediction']}")
-print(f"置信度: {result['confidence']:.2%}")
-```
-
-### 工具函数封装
-
-```python
-import requests
-import base64
-from typing import Dict, Any
-
-
-def classify_image_file(image_path: str, api_url: str = "http://localhost:5000") -> Dict[str, Any]:
-    """
-    分类图片文件（文件上传方式）
-
-    Args:
-        image_path: 图片文件路径
-        api_url: API服务地址
-
-    Returns:
-        分类结果字典
-    """
-    url = f"{api_url}/classify"
-    files = {"image": open(image_path, "rb")}
-
-    response = requests.post(url, files=files)
-    return response.json()
-
-
-def classify_image_base64(image_path: str, api_url: str = "http://localhost:5000") -> Dict[str, Any]:
-    """
-    分类图片文件（Base64方式）
-
-    Args:
-        image_path: 图片文件路径
-        api_url: API服务地址
-
-    Returns:
-        分类结果字典
-    """
-    # 读取并编码图片
-    with open(image_path, "rb") as f:
-        image_data = base64.b64encode(f.read()).decode("ascii")
-
-    url = f"{api_url}/classify_base64"
-    payload = {"image": image_data}
-
-    response = requests.post(url, json=payload)
-    return response.json()
-
-
-# 使用示例
-if __name__ == "__main__":
-    result = classify_image_file("test.jpg")
-
-    if result.get("success"):
-        print(f"✓ 预测结果: {result['prediction']}")
-        print(f"  置信度: {result['confidence']:.2%}")
-        print(f"  概率分布: {result['probabilities']}")
-    else:
-        print(f"✗ 错误: {result.get('error')}")
-```
-
-## 启动服务
-
-```bash
-cd printed-vs-handwritten
-python app.py
-```
-
-服务将在 `http://0.0.0.0:5000` 启动。
-
-## 集成到错题本工作流
-
-### 应用场景
-
-在错题本生成流程中，可以用手写体识别来：
-
-1. **题目分类优化**
-   - 识别题目区域是手写还是印刷
-   - 为手写题目提供不同的OCR参数或处理策略
-
-2. **答案区域检测**
-   - 检测学生是否已经手写作答
-   - 区分原始题目和学生答案
-
-3. **质量控制**
-   - 过滤掉手写笔记区域
-   - 只保留印刷体题目进行结构化处理
-
-### 集成方式
-
-可以作为 Agent 工具集成：
-
-```python
-from langchain_core.tools import tool
-import requests
-
-@tool
-def classify_text_type(image_path: str) -> str:
-    """
-    判断图片中的文字是手写还是印刷体
-
-    Args:
-        image_path: 图片路径
-
-    Returns:
-        "printed" 或 "handwritten"
-    """
-    try:
-        url = "http://localhost:5000/classify"
-        files = {"image": open(image_path, "rb")}
-
-        response = requests.post(url, files=files)
-        result = response.json()
-
-        if result.get("success"):
-            return f"{result['prediction']} (置信度: {result['confidence']:.2%})"
-        else:
-            return f"分类失败: {result.get('error')}"
-    except Exception as e:
-        return f"调用API出错: {str(e)}"
-```
-
-## 模型参数
-
-在代码中使用的模型参数：
-
-```python
-result = tgc.classify(
-    img,
-    stride=75,        # 滑动窗口步长
-    batch_size=64,    # 批处理大小
-    score_as_key=False
-)
-```
-
-这些参数可以根据实际需求调整以优化速度和精度。
-
-## 注意事项
-
-1. **图片格式**: 自动转换为 RGB 模式
-2. **模型加载**: 首次调用时加载模型（约需几秒）
-3. **并发**: Flask 默认单线程，高并发需要使用 gunicorn 等
-4. **内存**: 模型会常驻内存（约200-500MB）
-
-## 性能优化建议
-
-1. **生产部署**: 使用 gunicorn 代替 Flask 开发服务器
-   ```bash
-   gunicorn -w 4 -b 0.0.0.0:5000 app:app
-   ```
-
-2. **批量处理**: 如果需要处理多张图片，可以扩展 API 支持批量接口
-
-3. **缓存**: 对于相同图片可以缓存结果避免重复计算
diff --git a/docs/PROJECT_STRUCTURE.md b/docs/PROJECT_STRUCTURE.md
deleted file mode 100644
index d1864642..00000000
--- a/docs/PROJECT_STRUCTURE.md
+++ /dev/null
@@ -1,1359 +0,0 @@
-# 错题本生成系统 - 项目技术文档
-
-## 📋 项目概述
-
-### 系统简介
-这是一个基于 **LangChain Agent** 和 **PaddleOCR API** 的智能错题本生成系统。系统能够自动识别、分割和导出试卷/作业中的题目，生成结构化的错题本，支持选择题、填空题、解答题、判断题等多种题型。
-
-### 核心功能
-- ✅ PDF/图片文件自动标准化处理
-- ✅ 基于PaddleOCR的文档结构化解析（支持公式、图片、布局识别）
-- ✅ 基于DeepSeek的智能题目分割（AI自动识别题号、题型、选项）
-- ✅ 交互式HTML预览（可视化选择题目）
-- ✅ Markdown格式错题本导出
-
-### 技术栈
-- **LLM框架**: LangChain + LangGraph
-- **Agent**: DeepSeek Chat (via `create_deep_agent`)
-- **OCR**: PaddleOCR-VL API
-- **文档处理**: pdf2image, Pillow
-- **CLI**: Rich (美化输出)
-- **配置**: python-dotenv
-
----
-
-## 🗂️ 目录结构
-
-```
-error_correction/
-│
-├── error_correction_agent/          # 🤖 Agent智能层
-│   ├── agent.py                     # Agent工厂函数和配置
-│   ├── prompts.py                   # 系统提示词（题目分割规则）
-│   └── tools/                       # Agent工具集
-│       ├── __init__.py              # 工具统一导出
-│       ├── question_tools.py        # 题目处理工具
-│       └── file_tools.py            # 文件I/O工具
-│
-├── src/                             # ⚙️ 确定性业务逻辑层
-│   ├── paddleocr_client.py          # PaddleOCR API客户端
-│   ├── workflow.py                  # 5步工作流编排
-│   └── utils.py                     # 通用工具函数
-│
-├── network_search_agent/            # 🔍 网络搜索Agent（独立模块）
-│   ├── prompts.py
-│   └── tools.py
-│
-├── printed-vs-handwritten/          # ✍️ 手写/印刷识别子项目
-│   ├── app.py                       # Flask API服务（端口5000）
-│   ├── classify.py                  # 分类器CLI
-│   ├── train-model.py               # 模型训练
-│   └── ocrd_typegroups_classifier/  # 分类器核心库
-│
-├── input/                           # 📥 输入文件目录
-├── output/                          # 📤 输出目录
-│   ├── pages/                       # 标准化后的PNG图片
-│   ├── struct/                      # PaddleOCR解析结果（JSON）
-│   └── assets/                      # 下载的图片资源
-│
-├── results/                         # 🎯 最终结果目录
-│   ├── questions.json               # Agent分割后的题目
-│   ├── split_issues.jsonl           # 分割过程问题日志
-│   ├── preview.html                 # 交互式预览页面
-│   └── wrongbook.md                 # 最终错题本
-│
-├── docs/                            # 📚 文档
-│   ├── INDEX.md                     # LangChain文档索引
-│   ├── HANDWRITTEN_API.md           # 手写体识别API文档
-│   └── PROJECT_STRUCTURE.md         # 本文档
-│
-├── langgraph.json                   # LangGraph配置文件
-├── .env                             # 环境变量配置（不提交）
-├── .env.example                     # 配置模板
-├── README.md                        # 快速开始指南
-├── requirements.txt                 # Python依赖
-├── test_paddleocr.py                # PaddleOCR测试脚本
-├── test_workflow.py                 # 工作流测试脚本
-└── agent.py                         # network_search_agent入口
-```
-
----
-
-## 🔧 核心模块详解
-
-### 1. error_correction_agent/ - Agent智能层
-
-#### 📄 agent.py
-
-**作用**: 创建和配置题目分割Agent
-
-**核心函数**:
-
-```python
-def create_question_split_agent() -> Agent:
-    """
-    创建题目分割Agent
-
-    使用DeepSeek模型和定义的工具创建一个专门用于分割题目的Agent。
-
-    Returns:
-        配置好的Agent实例
-    """
-    # 初始化DeepSeek模型
-    model = init_chat_model(
-        model="deepseek:deepseek-chat",
-        temperature=0.1,  # 低温度确保稳定输出
-    )
-
-    # 绑定4个工具
-    tools = [
-        save_questions,
-        log_issue,
-        download_image,
-        read_ocr_result,
-    ]
-
-    # 创建Agent
-    agent = create_deep_agent(
-        model=model,
-        tools=tools,
-        system_prompt=SYSTEM_PROMPT,
-    )
-
-    return agent
-```
-
-**全局导出**:
-```python
-agent = create_question_split_agent()  # 供langgraph.json使用
-```
-
-**依赖关系**:
-```
-create_question_split_agent()
-├─→ init_chat_model()           [langchain.chat_models]
-├─→ create_deep_agent()         [deepagents]
-├─→ SYSTEM_PROMPT               [.prompts]
-└─→ 4 Tools                     [.tools]
-    ├─→ save_questions
-    ├─→ log_issue
-    ├─→ download_image
-    └─→ read_ocr_result
-```
-
----
-
-#### 📄 prompts.py
-
-**作用**: 定义Agent的系统提示词，指导题目分割行为
-
-**核心内容**:
-
-```python
-SYSTEM_PROMPT = """# 错题本题目分割专家
-
-你是一个专业的试卷题目分割专家。你的任务是分析OCR识别的文档结构，智能分割出独立的题目。
-
-## 核心任务
-从PaddleOCR解析的结构化数据中：
-1. 识别题目边界（题号是关键标志）
-2. 提取题干内容
-3. 识别并提取选项（如果是选择题）
-4. 识别公式块（display_formula 和 inline_formula）
-5. 识别图片块并保留引用
-6. 按阅读顺序组织内容
-
-## 题目类型识别
-- 选择题: 包含A/B/C/D选项
-- 填空题: 包含下划线或括号
-- 解答题: 要求"解答"、"证明"、"计算"
-- 判断题: 要求判断"正确/错误"
-
-## 题号识别规则
-- 1. / 2. / 3. ...
-- 1、/ 2、/ 3、...
-- （1）/（2）/（3）...
-- 一、/ 二、/ 三、...
-
-## 工作流程
-1. 分析block顺序（使用block_order）
-2. 识别题目起点（查找题号）
-3. 收集题目内容（直到下一个题号）
-4. 结构化输出
-
-...（详见完整提示词）
-"""
-```
-
-**输出格式规范**:
-```json
-{
-  "question_id": "1",
-  "question_type": "选择题",
-  "content_blocks": [
-    {
-      "block_type": "text|display_formula|inline_formula|image",
-      "content": "内容文本",
-      "bbox": [x1, y1, x2, y2],
-      "block_id": 原始ID
-    }
-  ],
-  "options": ["A. ...", "B. ...", "C. ...", "D. ..."],
-  "has_formula": true,
-  "has_image": false,
-  "image_refs": ["imgs/xxx.jpg"]
-}
-```
-
----
-
-#### 📄 tools/question_tools.py
-
-**工具1: save_questions**
-
-```python
-@tool(parse_docstring=True)
-def save_questions(
-    questions: List[Dict[str, Any]],
-    output_path: str = None
-) -> str:
-    """
-    保存分割后的题目列表到JSON文件
-
-    Args:
-        questions: 题目列表，每个题目包含:
-            - question_id: 题号
-            - question_type: 题型
-            - content_blocks: 内容块列表
-            - options: 选项（选择题）
-            - has_formula: 是否包含公式
-            - has_image: 是否包含图片
-            - image_refs: 图片引用路径列表
-        output_path: 输出路径（默认: {RESULTS_DIR}/questions.json）
-
-    Returns:
-        保存结果消息
-    """
-```
-
-**使用场景**: Agent分割完成后，调用此工具保存题目列表
-
-**工具2: log_issue**
-
-```python
-@tool(parse_docstring=True)
-def log_issue(
-    issue_type: str,
-    description: str,
-    block_info: Dict[str, Any] = None
-) -> str:
-    """
-    记录题目分割过程中遇到的问题
-
-    Args:
-        issue_type: 问题类型
-            - "unclear_boundary": 边界不清
-            - "missing_question_number": 缺少题号
-            - "complex_structure": 复杂结构
-        description: 问题详细描述
-        block_info: 相关block信息（可选）
-
-    Returns:
-        记录结果消息
-    """
-```
-
-**输出**: 追加到 `{RESULTS_DIR}/split_issues.jsonl`
-
-**使用场景**: Agent遇到不确定情况时记录，供人工审核或算法改进
-
----
-
-#### 📄 tools/file_tools.py
-
-**工具3: download_image**
-
-```python
-@tool(parse_docstring=True)
-def download_image(image_url: str, save_path: str) -> str:
-    """
-    从URL下载图片到本地
-
-    Args:
-        image_url: 图片URL地址
-        save_path: 保存路径（相对于ASSETS_DIR）
-
-    Returns:
-        下载结果消息
-    """
-```
-
-**使用场景**: 下载PaddleOCR返回的图片资源
-
-**工具4: read_ocr_result**
-
-```python
-@tool(parse_docstring=True)
-def read_ocr_result(result_path: str) -> Dict[str, Any]:
-    """
-    读取PaddleOCR的解析结果
-
-    Args:
-        result_path: OCR结果文件路径（JSON）
-
-    Returns:
-        OCR解析结果字典或错误信息
-    """
-```
-
-**使用场景**: Agent需要重新读取或分析OCR结果时
-
----
-
-### 2. src/ - 确定性业务逻辑层
-
-#### 📄 paddleocr_client.py
-
-**类: PaddleOCRClient**
-
-**初始化方法**:
-```python
-def __init__(self):
-    """
-    初始化客户端，从环境变量读取配置
-
-    环境变量:
-        PADDLEOCR_API_URL: API端点
-        PADDLEOCR_API_TOKEN: 认证令牌
-        PADDLEOCR_USE_DOC_ORIENTATION: 启用方向检测
-        PADDLEOCR_USE_DOC_UNWARPING: 启用文档展平
-        PADDLEOCR_USE_CHART_RECOGNITION: 启用图表识别
-    """
-```
-
-**主要方法**:
-
-**1️⃣ parse_image**
-```python
-def parse_image(
-    self,
-    image_path: str,
-    save_output: bool = True,
-    output_dir: Optional[str] = None
-) -> Dict[str, Any]:
-    """
-    解析图片并返回结构化结果
-
-    流程:
-        1. 读取图片并转为Base64
-        2. 构建API请求payload
-        3. 调用PaddleOCR API
-        4. 保存JSON结果到 {STRUCT_DIR}/{filename}_struct.json
-        5. 下载并保存响应中的图片和Markdown
-
-    Args:
-        image_path: 图片文件路径
-        save_output: 是否保存输出（默认True）
-        output_dir: 输出目录（默认从STRUCT_DIR读取）
-
-    Returns:
-        PaddleOCR API返回的结构化结果
-    """
-```
-
-**返回结构示例**:
-```python
-{
-    "layoutParsingResults": [
-        {
-            "prunedResult": {
-                "parsing_res_list": [
-                    {
-                        "block_id": 1,
-                        "block_type": "text",
-                        "block_label": "paragraph_title",
-                        "block_content": "1. 题目内容...",
-                        "block_bbox": [x1, y1, x2, y2],
-                        "block_order": 1,
-                        "group_id": 1
-                    },
-                    {
-                        "block_type": "display_formula",
-                        "block_content": "x^2 + y^2 = r^2",
-                        ...
-                    }
-                ]
-            },
-            "block_order": [1, 2, 3, 4, ...],  # 推荐阅读顺序
-            "markdown": {
-                "text": "## 第一题\n\n1. ...",
-                "images": {
-                    "imgs/figure_1.jpg": "https://..."
-                }
-            },
-            "outputImages": {
-                "layout": "https://...",
-                "table": "https://..."
-            }
-        }
-    ]
-}
-```
-
-**2️⃣ parse_pdf**
-```python
-def parse_pdf(
-    self,
-    pdf_path: str,
-    save_output: bool = True,
-    output_dir: Optional[str] = None
-) -> Dict[str, Any]:
-    """
-    解析PDF并返回结构化结果
-
-    功能同parse_image，但处理PDF文件
-    """
-```
-
-**3️⃣ _save_images**
-```python
-def _save_images(self, result: Dict[str, Any], output_dir: str):
-    """
-    下载并保存结果中的图片资源
-
-    保存内容:
-        1. Markdown文档 (doc_0.md, doc_1.md, ...)
-        2. Markdown中的图片
-        3. outputImages中的可视化图片
-    """
-```
-
----
-
-#### 📄 workflow.py
-
-**类: ErrorCorrectionWorkflow**
-
-**作用**: 编排5步完整工作流
-
-**初始化**:
-```python
-def __init__(self):
-    """
-    初始化工作流
-
-    属性:
-        paddleocr_client: PaddleOCR客户端实例
-        image_paths: 步骤1输出（标准化图片路径）
-        ocr_results: 步骤2输出（OCR解析结果）
-        questions: 步骤3输出（Agent分割的题目）
-    """
-```
-
-**5步工作流**:
-
-| 步骤 | 方法 | 类型 | 输入 | 输出 |
-|-----|------|------|------|------|
-| 1 | `prepare_input` | 确定性 | `file_path: str` | `image_paths: List[str]` |
-| 2 | `paddleocr_parse` | 确定性 | `image_paths: List[str]` | `ocr_results: List[Dict]` |
-| 3 | `split_questions_with_agent` | 🤖智能 | `ocr_results: List[Dict]` | `questions: List[Dict]` |
-| 4 | `build_preview` | 确定性 | `questions: List[Dict]` | `preview.html` |
-| 5 | `export_selected` | 确定性 | `questions, selected_ids` | `wrongbook.md` |
-
-**核心方法详解**:
-
-**主入口: run**
-```python
-def run(self, input_file: str, auto_split: bool = False) -> Dict[str, Any]:
-    """
-    运行完整工作流
-
-    Args:
-        input_file: 输入文件路径（PDF或图片）
-        auto_split: 是否自动调用Agent分割（默认False）
-
-    Returns:
-        包含各步骤结果的字典:
-            - image_paths: 标准化图片路径列表
-            - ocr_results: OCR解析结果列表
-            - questions: 分割后的题目列表
-
-    执行步骤:
-        1. 准备输入文件（调用utils.prepare_input）
-        2. PaddleOCR解析（调用self.paddleocr_parse）
-        3. (可选) Agent分割题目
-        4. (如有题目) 生成HTML预览
-    """
-```
-
-**步骤2实现**:
-```python
-def paddleocr_parse(self, image_paths: List[str]) -> List[Dict[str, Any]]:
-    """
-    调用PaddleOCR解析文档结构
-
-    Args:
-        image_paths: 图片路径列表
-
-    Returns:
-        每张图片的PaddleOCR解析结果列表
-
-    过程:
-        对每张图片调用 PaddleOCRClient.parse_image()
-        自动保存JSON和图片资源
-    """
-```
-
-**步骤3实现**（核心智能步骤）:
-```python
-def split_questions_with_agent(self) -> List[Dict[str, Any]]:
-    """
-    使用Agent智能分割题目
-
-    Returns:
-        分割后的题目列表
-
-    执行流程:
-        1. 创建Agent实例
-        2. 简化OCR结果为Agent友好格式
-        3. 构建提示词
-        4. 调用Agent.invoke()
-        5. 从 {RESULTS_DIR}/questions.json 读取Agent保存的结果
-
-    简化格式:
-        simplified_results = [
-            {
-                "blocks": [...],        # parsing_res_list
-                "block_order": [...]    # 推荐阅读顺序
-            }
-        ]
-    """
-```
-
-**步骤5实现**:
-```python
-def export_selected(
-    self,
-    selected_ids: List[str],
-    output_path: str = None
-) -> str:
-    """
-    导出选中的题目为错题本
-
-    Args:
-        selected_ids: 用户选择的题目ID列表
-        output_path: 输出路径（可选）
-
-    Returns:
-        导出文件路径
-
-    过程:
-        1. 过滤选中题目
-        2. 调用 utils.export_wrongbook()
-        3. 返回Markdown文件路径
-    """
-```
-
----
-
-#### 📄 utils.py
-
-**函数1: prepare_input (步骤1)**
-
-```python
-def prepare_input(file_path: str) -> List[str]:
-    """
-    准备输入文件 - 统一转换为标准PNG图片
-
-    Args:
-        file_path: 输入文件路径（PDF或图片）
-
-    Returns:
-        标准化后的PNG图片路径列表
-
-    处理逻辑:
-        如果是PDF:
-            1. 使用pdf2image转换为300dpi的图片
-            2. 逐页保存为PNG格式
-            3. 命名: {stem}_page_001.png, {stem}_page_002.png, ...
-
-        如果是图片:
-            1. 打开图片
-            2. 转换为RGB模式
-            3. 保存为PNG格式
-            4. 命名: {stem}.png
-
-    输出目录: {PAGES_DIR}
-
-    支持格式:
-        - PDF: .pdf
-        - 图片: .jpg, .jpeg, .png, .bmp, .tiff, .webp
-
-    Raises:
-        FileNotFoundError: 文件不存在
-        ValueError: 不支持的文件格式
-    """
-```
-
-**函数2: build_preview (步骤4)**
-
-```python
-def build_preview(
-    questions: List[Dict[str, Any]],
-    output_path: str = None
-) -> str:
-    """
-    生成HTML预览页面
-
-    Args:
-        questions: 题目列表
-        output_path: 输出路径（默认: {RESULTS_DIR}/preview.html）
-
-    Returns:
-        HTML文件路径
-
-    页面功能:
-        1. 展示所有题目（题号、题型、内容）
-        2. 支持点击选择题目
-        3. 选中状态可视化（按钮变色）
-        4. 底部导出按钮（显示选中题目ID）
-
-    样式特点:
-        - 题目卡片布局
-        - 题型色标（选择题=绿色）
-        - 公式高亮显示
-        - 图片引用标记
-        - 响应式设计
-    """
-```
-
-**函数3: export_wrongbook (步骤5)**
-
-```python
-def export_wrongbook(
-    questions: List[Dict[str, Any]],
-    selected_ids: List[str],
-    output_path: str = None
-) -> str:
-    """
-    导出错题本为Markdown格式
-
-    Args:
-        questions: 题目列表
-        selected_ids: 选中的题目ID列表
-        output_path: 输出路径（默认: {RESULTS_DIR}/wrongbook.md）
-
-    Returns:
-        Markdown文件路径
-
-    Markdown格式:
-        # 错题本
-        > 共收录 N 道题目
-        ---
-
-        ## 1. 题目 X (题型)
-
-        [题干内容]
-        [选项或公式]
-        [图片引用]
-
-        ### 我的答案
-        _（请在此处填写你的答案）_
-
-        ### 正确答案
-        _（请在此处填写正确答案）_
-
-        ### 解析
-        _（请在此处填写解题思路和知识点）_
-
-        ---
-
-    特性:
-        - 支持LaTeX公式（$...$和$$...$$）
-        - 图片Markdown引用
-        - 预留答案和解析填写区域
-    """
-```
-
----
-
-## 🔗 模块依赖关系图
-
-```
-┌─────────────────────────────────────────────────────────────┐
-│                    User Entry Point                         │
-│              ErrorCorrectionWorkflow.run()                   │
-└──────────────┬──────────────────────────────────────────────┘
-               │
-               ├── [步骤1] prepare_input(file_path)
-               │            │
-               │            ├─ PDF? → pdf2image.convert_from_path()
-               │            └─ 图片? → PIL.Image.open() → save PNG
-               │            │
-               │            └→ List[image_paths]
-               │
-               ├── [步骤2] paddleocr_parse(image_paths)
-               │            │
-               │            └─ for each image:
-               │                ├─ PaddleOCRClient.parse_image()
-               │                │   ├─ base64编码
-               │                │   ├─ requests.post(API_URL)
-               │                │   ├─ 保存JSON结果
-               │                │   └─ _save_images()
-               │                │       ├─ 下载markdown图片
-               │                │       ├─ 下载output图片
-               │                │       └─ 保存markdown文档
-               │                │
-               │                └→ List[ocr_results]
-               │
-               ├── [步骤3] split_questions_with_agent() 🤖
-               │            │
-               │            ├─ create_question_split_agent()
-               │            │   ├─ init_chat_model("deepseek:deepseek-chat")
-               │            │   ├─ tools = [save_questions, log_issue, ...]
-               │            │   └─ create_deep_agent(model, tools, SYSTEM_PROMPT)
-               │            │
-               │            ├─ 简化OCR结果:
-               │            │   simplified_results = [
-               │            │       {"blocks": [...], "block_order": [...]}
-               │            │   ]
-               │            │
-               │            ├─ 构建提示词
-               │            │
-               │            ├─ agent.invoke({
-               │            │       "messages": [
-               │            │           {"role": "user", "content": prompt},
-               │            │           {"role": "user", "content": OCR数据}
-               │            │       ]
-               │            │   })
-               │            │   │
-               │            │   └─ Agent内部流程:
-               │            │       ├─ 分析题号和内容结构
-               │            │       ├─ 识别题型
-               │            │       ├─ 可能调用: download_image()
-               │            │       ├─ 可能调用: log_issue()
-               │            │       └─ 调用: save_questions()
-               │            │           └─ 保存到 results/questions.json
-               │            │
-               │            └─ 读取 questions.json → List[questions]
-               │
-               ├── [步骤4] build_preview(questions)
-               │            │
-               │            ├─ 构建HTML内容
-               │            │   ├─ 题目卡片
-               │            │   ├─ 选择按钮
-               │            │   └─ JavaScript交互
-               │            │
-               │            └─ 保存到 results/preview.html
-               │
-               └── [步骤5] export_selected(questions, selected_ids)
-                            │
-                            ├─ 过滤选中题目
-                            ├─ 生成Markdown内容
-                            │   ├─ 题干
-                            │   ├─ 选项/公式
-                            │   ├─ 答案区域（空白）
-                            │   └─ 解析区域（空白）
-                            │
-                            └─ 保存到 results/wrongbook.md
-```
-
----
-
-## ⚙️ 环境变量配置表
-
-| 变量 | 必需? | 用途 | 示例值 | 默认值 |
-|-----|------|------|--------|--------|
-| **PaddleOCR配置** | | | | |
-| `PADDLEOCR_API_URL` | ✅ | PaddleOCR API端点 | `https://api.xxx.com/layout-parsing` | - |
-| `PADDLEOCR_API_TOKEN` | ✅ | API认证令牌 | `token_abc123` | - |
-| `PADDLEOCR_USE_DOC_ORIENTATION` | ❌ | 启用文档方向检测 | `true` / `false` | `false` |
-| `PADDLEOCR_USE_DOC_UNWARPING` | ❌ | 启用文档展平 | `true` / `false` | `false` |
-| `PADDLEOCR_USE_CHART_RECOGNITION` | ❌ | 启用图表识别 | `true` / `false` | `false` |
-| **DeepSeek配置** | | | | |
-| `DEEPSEEK_API_KEY` | ✅ | DeepSeek API密钥 | `sk-xxx` | - |
-| `DEEPSEEK_BASE_URL` | ❌ | DeepSeek基础URL | `https://api.deepseek.com` | 自动 |
-| **LangSmith配置** | | | | |
-| `LANGSMITH_TRACING` | ❌ | 启用LangSmith追踪 | `true` / `false` | `false` |
-| `LANGSMITH_API_KEY` | ❌ | LangSmith认证密钥 | `lsv2_pt_xxx` | - |
-| `LANGSMITH_PROJECT` | ❌ | LangSmith项目名称 | `error-correction` | - |
-| `LANGSMITH_ENDPOINT` | ❌ | LangSmith端点 | `https://api.smith.langchain.com` | 默认 |
-| **其他API** | | | | |
-| `TAVILY_API_KEY` | ❌ | Tavily搜索API密钥 | `tvly-xxx` | - |
-| `ANTHROPIC_API_KEY` | ❌ | Anthropic API密钥 | `sk-ant-xxx` | - |
-| **输出目录配置** | | | | |
-| `OUTPUT_DIR` | ❌ | 总输出目录 | `output` | `output` |
-| `PAGES_DIR` | ❌ | 标准化图片目录 | `output/pages` | `output/pages` |
-| `STRUCT_DIR` | ❌ | OCR结果目录 | `output/struct` | `output/struct` |
-| `ASSETS_DIR` | ❌ | 资源目录 | `output/assets` | `output/assets` |
-| `RESULTS_DIR` | ❌ | 最终结果目录 | `results` | `results` |
-
----
-
-## 💻 使用流程示例
-
-### 方式1: 分步骤手动控制
-
-```python
-from src.workflow import ErrorCorrectionWorkflow
-
-# 创建工作流实例
-workflow = ErrorCorrectionWorkflow()
-
-# 步骤1-2: 准备输入和OCR解析
-result = workflow.run("input/exam_paper.pdf", auto_split=False)
-
-print(f"已处理 {len(result['image_paths'])} 张图片")
-print(f"OCR解析完成，结果保存在 output/struct/")
-
-# 步骤3: 手动调用Agent分割题目
-questions = workflow.split_questions_with_agent()
-
-print(f"分割出 {len(questions)} 道题目")
-
-# 步骤4: 自动生成HTML预览
-# (在run()中已自动调用)
-
-# 步骤5: 导出选中的题目
-selected_ids = ["1", "2", "5", "7"]
-wrongbook_path = workflow.export_selected(selected_ids)
-
-print(f"错题本已导出: {wrongbook_path}")
-```
-
-### 方式2: 自动完整流程
-
-```python
-from src.workflow import ErrorCorrectionWorkflow
-
-# 创建工作流实例
-workflow = ErrorCorrectionWorkflow()
-
-# 自动执行步骤1-4
-workflow.run("input/exam_paper.pdf", auto_split=True)
-
-# 打开 results/preview.html 查看题目
-# 选择题目后，手动导出
-workflow.export_selected(["1", "3", "5"])
-```
-
-### 方式3: 仅测试PaddleOCR
-
-```python
-from src.paddleocr_client import PaddleOCRClient
-
-client = PaddleOCRClient()
-
-# 解析单张图片
-result = client.parse_image("input/page1.jpg")
-
-# 查看结果
-print(f"识别的块数量: {len(result['layoutParsingResults'][0]['prunedResult']['parsing_res_list'])}")
-
-# 结果自动保存到 output/struct/page1_struct.json
-```
-
-### 方式4: 使用测试脚本
-
-```bash
-# 测试步骤1-2（不调用Agent）
-python test_workflow.py
-
-# 测试完整流程（包含Agent）
-python test_workflow.py --with-agent
-
-# 测试PaddleOCR
-python test_paddleocr.py
-```
-
----
-
-## 🤖 Agent执行流程详解
-
-### Agent接收的数据格式
-
-```python
-simplified_results = [
-    {
-        "blocks": [
-            {
-                "block_id": 1,
-                "block_type": "text",
-                "block_label": "paragraph_title",
-                "block_content": "1. 下列关于平行四边形的说法，正确的是（  ）",
-                "block_bbox": [120, 300, 800, 350],
-                "block_order": 1,
-                "group_id": 1
-            },
-            {
-                "block_id": 2,
-                "block_type": "text",
-                "block_label": "text",
-                "block_content": "A. 对边相等",
-                "block_bbox": [150, 360, 400, 390],
-                "block_order": 2,
-                "group_id": 1
-            },
-            {
-                "block_id": 3,
-                "block_type": "display_formula",
-                "block_content": "\\sum_{i=1}^{n} x_i = S",
-                "block_bbox": [200, 400, 600, 450],
-                "block_order": 3,
-                "group_id": 1
-            }
-        ],
-        "block_order": [1, 2, 3, 4, 5, ...]
-    }
-]
-```
-
-### Agent的思考过程
-
-```
-1️⃣ 接收OCR数据
-   ↓
-2️⃣ 分析block_order，按推荐顺序扫描
-   ↓
-3️⃣ 识别题号模式
-   - 检测 "1." / "1、" / "（1）" 等
-   - 记录题目起始位置
-   ↓
-4️⃣ 收集题目内容
-   - 从题号开始收集后续block
-   - 遇到下一个题号停止
-   - 包括text、formula、image等
-   ↓
-5️⃣ 识别题型
-   - 是否包含选项 (A/B/C/D) → 选择题
-   - 是否包含下划线 → 填空题
-   - 是否要求"解答" → 解答题
-   ↓
-6️⃣ 提取选项（如果是选择题）
-   - 识别A. B. C. D.模式
-   - 或A) B) C) D)模式
-   ↓
-7️⃣ 处理特殊内容
-   - 公式: has_formula = true
-   - 图片: has_image = true, image_refs = [...]
-   ↓
-8️⃣ 可能的工具调用
-   - download_image() - 如需保存图片
-   - log_issue() - 如遇到不确定情况
-   ↓
-9️⃣ 结构化输出
-   ↓
-🔟 调用 save_questions() 保存结果
-```
-
-### Agent工具调用时机
-
-| 工具 | 调用时机 | 示例场景 |
-|-----|---------|---------|
-| `save_questions` | 必须 | 分割完成后保存题目列表 |
-| `log_issue` | 可选 | 题号识别不确定时记录 |
-| `download_image` | 可选 | OCR结果包含图片URL时 |
-| `read_ocr_result` | 可选 | 需要重新分析OCR数据时 |
-
-### Agent执行日志示例
-
-```
-[Agent] 正在分析OCR数据...
-[Agent] 识别到题号: "1."
-[Agent] 题型判断: 选择题 (检测到选项A、B、C、D)
-[Agent] 收集内容块: block_1, block_2, block_3, block_4
-[Agent] 检测到公式块: display_formula
-[Tool] 调用 save_questions: 保存1道题目
-[Agent] 识别到题号: "2."
-[Agent] 题型判断: 填空题 (检测到下划线)
-[Tool] 调用 log_issue: 题号2的边界不清晰
-[Tool] 调用 save_questions: 保存2道题目
-[Agent] 分割完成，共处理2道题目
-```
-
----
-
-## 🔧 扩展性设计
-
-### 1. 添加新Agent工具
-
-#### 步骤1: 创建工具文件
-
-在 `error_correction_agent/tools/` 创建新文件（如 `image_tools.py`）:
-
-```python
-from langchain_core.tools import tool
-from PIL import Image
-
-@tool(parse_docstring=True)
-def crop_question_region(image_path: str, bbox: list) -> str:
-    """
-    根据bbox坐标裁剪题目区域
-
-    Args:
-        image_path: 原始图片路径
-        bbox: 边界框坐标 [x1, y1, x2, y2]
-
-    Returns:
-        裁剪后的图片路径
-    """
-    img = Image.open(image_path)
-    cropped = img.crop(tuple(bbox))
-
-    output_path = f"output/assets/cropped_{os.path.basename(image_path)}"
-    cropped.save(output_path)
-
-    return f"已裁剪并保存到: {output_path}"
-```
-
-#### 步骤2: 导出工具
-
-在 `error_correction_agent/tools/__init__.py` 添加:
-
-```python
-from .image_tools import crop_question_region
-
-__all__ = [
-    "save_questions",
-    "log_issue",
-    "download_image",
-    "read_ocr_result",
-    "crop_question_region",  # 新增
-]
-```
-
-#### 步骤3: 绑定到Agent
-
-在 `error_correction_agent/agent.py` 添加:
-
-```python
-from .tools import (
-    save_questions,
-    log_issue,
-    download_image,
-    read_ocr_result,
-    crop_question_region  # 新增
-)
-
-def create_question_split_agent():
-    tools = [
-        save_questions,
-        log_issue,
-        download_image,
-        read_ocr_result,
-        crop_question_region,  # 新增
-    ]
-    ...
-```
-
-### 2. 自定义工作流
-
-#### 在现有流程中插入步骤
-
-编辑 `src/workflow.py`:
-
-```python
-class ErrorCorrectionWorkflow:
-
-    def run(self, input_file: str, auto_split: bool = False):
-        # 原有步骤
-        self.image_paths = prepare_input(input_file)
-        self.ocr_results = self.paddleocr_parse(self.image_paths)
-
-        # 新增步骤: 手写体识别
-        self.classify_handwritten()
-
-        # 继续原有流程
-        if auto_split:
-            self.questions = self.split_questions_auto()
-        ...
-
-    def classify_handwritten(self):
-        """新步骤: 识别手写/印刷体"""
-        import requests
-
-        for image_path in self.image_paths:
-            files = {"image": open(image_path, "rb")}
-            response = requests.post(
-                "http://localhost:5000/classify",
-                files=files
-            )
-            result = response.json()
-
-            if result["prediction"] == "handwritten":
-                console.print(f"[yellow]检测到手写内容: {image_path}[/yellow]")
-```
-
-#### 创建新工作流变体
-
-```python
-class InteractiveWorkflow(ErrorCorrectionWorkflow):
-    """交互式工作流，每步都请求确认"""
-
-    def run(self, input_file: str):
-        # 步骤1
-        self.image_paths = prepare_input(input_file)
-        input("按Enter继续步骤2...")
-
-        # 步骤2
-        self.ocr_results = self.paddleocr_parse(self.image_paths)
-        input("按Enter继续步骤3...")
-
-        # 步骤3
-        self.questions = self.split_questions_with_agent()
-        ...
-```
-
-### 3. 调整Agent行为
-
-#### 修改提示词
-
-编辑 `error_correction_agent/prompts.py`:
-
-```python
-SYSTEM_PROMPT = """# 错题本题目分割专家
-
-...（原有内容）
-
-## 新增规则
-
-### 识别小题
-如果题目包含子题（如1. (1) (2) (3)），请拆分为独立题目：
-- question_id: "1-1", "1-2", "1-3"
-- parent_id: "1"
-
-### 数学题型细分
-对于解答题，进一步识别：
-- "计算题": 要求计算具体数值
-- "证明题": 要求证明结论
-- "应用题": 实际问题情景
-"""
-```
-
-#### 调整模型参数
-
-编辑 `error_correction_agent/agent.py`:
-
-```python
-def create_question_split_agent():
-    model = init_chat_model(
-        model="deepseek:deepseek-chat",
-        temperature=0.3,  # 提高温度以增加创造性
-        # max_tokens=4000,  # 限制输出长度
-    )
-    ...
-```
-
-#### 添加Few-shot示例
-
-在提示词中添加示例：
-
-```python
-SYSTEM_PROMPT = """
-...（原有内容）
-
-## 示例
-
-### 示例1: 选择题
-输入:
-{
-  "blocks": [
-    {"content": "1. 下列说法正确的是（  ）", ...},
-    {"content": "A. 选项1", ...},
-    {"content": "B. 选项2", ...}
-  ]
-}
-
-输出:
-{
-  "question_id": "1",
-  "question_type": "选择题",
-  "options": ["A. 选项1", "B. 选项2"],
-  ...
-}
-
-### 示例2: 填空题
-...
-"""
-```
-
----
-
-## 📊 技术指标
-
-### 代码规模
-
-| 文件 | 行数 | 主要功能 | 复杂度 |
-|-----|------|---------|--------|
-| `agent.py` | 56 | Agent创建 | ⭐ |
-| `prompts.py` | 133 | 系统提示词 | ⭐⭐ |
-| `question_tools.py` | 82 | 题目工具 | ⭐⭐ |
-| `file_tools.py` | 68 | 文件工具 | ⭐ |
-| `paddleocr_client.py` | 262 | PaddleOCR客户端 | ⭐⭐⭐ |
-| `workflow.py` | 214 | 工作流编排 | ⭐⭐⭐⭐ |
-| `utils.py` | 365 | 工具函数 | ⭐⭐⭐ |
-| **总计** | **1180** | | |
-
-### 核心依赖包
-
-```txt
-# Agent框架
-deepagents==0.3.5
-langchain==1.2.3
-langchain-core==1.2.7
-langgraph==1.0.5
-langchain-deepseek==1.0.1
-
-# 文档处理
-pdf2image>=1.16.3
-Pillow>=10.0.0
-
-# OCR
-# (通过API调用，无需本地依赖)
-
-# 配置和工具
-python-dotenv==1.2.1
-requests>=2.31.0
-rich==14.2.0
-pydantic>=2.0.0
-
-# 可选: 手写体识别
-torch>=2.0.0
-torchvision>=0.15.0
-```
-
-### 性能指标
-
-| 步骤 | 平均耗时 | 依赖 |
-|-----|---------|------|
-| 1. 准备输入 | ~2-5秒/PDF页 | PDF转图片 |
-| 2. OCR解析 | ~5-10秒/页 | PaddleOCR API |
-| 3. Agent分割 | ~10-30秒/页 | DeepSeek API |
-| 4. 生成预览 | <1秒 | 本地HTML生成 |
-| 5. 导出错题本 | <1秒 | 本地Markdown生成 |
-
-**总耗时估算**: 约20-50秒/页（主要取决于API响应速度）
-
----
-
-## 🔍 常见问题
-
-### Q1: Agent分割效果不好怎么办？
-
-**方案1**: 调整提示词
-- 编辑 `prompts.py`，添加更详细的规则
-- 添加few-shot示例
-
-**方案2**: 调整温度参数
-- 在 `agent.py` 中修改 `temperature`
-- 建议范围: 0.1-0.3
-
-**方案3**: 查看问题日志
-- 检查 `results/split_issues.jsonl`
-- 分析Agent记录的不确定情况
-
-### Q2: 如何调试Agent执行过程？
-
-**方法1**: 启用LangSmith追踪
-```bash
-# 在.env中设置
-LANGSMITH_TRACING=true
-LANGSMITH_API_KEY=your_key
-LANGSMITH_PROJECT=error-correction
-```
-
-**方法2**: 添加日志输出
-```python
-# 在workflow.py的split_questions_with_agent中
-response = agent.invoke(...)
-print(f"Agent响应: {response}")
-```
-
-### Q3: 如何处理大文件？
-
-**方案1**: 分批处理
-```python
-workflow = ErrorCorrectionWorkflow()
-
-# 每次只处理几页
-for i in range(0, len(pdf_pages), 5):
-    batch = pdf_pages[i:i+5]
-    workflow.run(batch, auto_split=True)
-```
-
-**方案2**: 调整API参数
-- 减少DPI（默认300）
-- 关闭部分OCR功能开关
-
-### Q4: 如何集成手写体识别？
-
-参考 `docs/HANDWRITTEN_API.md`：
-
-1. 启动服务: `cd printed-vs-handwritten && python app.py`
-2. 添加工具到Agent
-3. 在工作流中调用分类
-
----
-
-## 📝 开发规范
-
-### 代码风格
-- 遵循PEP 8
-- 使用类型提示（Type Hints）
-- 函数必须包含Docstring
-
-### 工具定义规范
-```python
-@tool(parse_docstring=True)  # 必须使用parse_docstring
-def tool_name(param: type) -> return_type:
-    """
-    简短描述（一句话）
-
-    详细说明（可选，多段落）
-
-    Args:
-        param: 参数说明
-
-    Returns:
-        返回值说明
-    """
-```
-
-### 提交规范
-- 使用语义化提交消息
-- 示例: `feat: 添加手写体识别工具`
-- 类型: `feat`, `fix`, `docs`, `refactor`, `test`
-
----
-
-## 🎯 未来计划
-
-### 短期（v1.1）
-- [ ] 添加手写体识别集成
-- [ ] 优化Agent提示词（few-shot示例）
-- [ ] 完善错误处理和日志
-
-### 中期（v1.2）
-- [ ] 支持批量处理多个文件
-- [ ] Web UI界面
-- [ ] 题目难度自动分级
-
-### 长期（v2.0）
-- [ ] 支持更多题型（完形填空、阅读理解）
-- [ ] 答案自动校对
-- [ ] 知识点自动标注
-
----
-
-## 📚 相关文档
-
-- [README.md](../README.md) - 快速开始指南
-- [INDEX.md](INDEX.md) - LangChain文档索引
-- [HANDWRITTEN_API.md](HANDWRITTEN_API.md) - 手写体识别API
-- [.env.example](../.env.example) - 环境配置模板
-
----
-
-**文档版本**: v1.0.0
-**最后更新**: 2026-01-27
-**维护者**: Claude Code Assistant
diff --git "a/example_uploads/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\345\214\226\345\255\246\347\234\237\351\242\230\357\274\210B\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf" "b/example_uploads/exams/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\345\214\226\345\255\246\347\234\237\351\242\230\357\274\210B\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
similarity index 100%
rename from "example_uploads/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\345\214\226\345\255\246\347\234\237\351\242\230\357\274\210B\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
rename to "example_uploads/exams/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\345\214\226\345\255\246\347\234\237\351\242\230\357\274\210B\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
diff --git "a/example_uploads/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\347\211\251\347\220\206\350\257\225\351\242\230\357\274\210A\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf" "b/example_uploads/exams/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\347\211\251\347\220\206\350\257\225\351\242\230\357\274\210A\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
similarity index 100%
rename from "example_uploads/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\347\211\251\347\220\206\350\257\225\351\242\230\357\274\210A\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
rename to "example_uploads/exams/2023\345\271\264\351\207\215\345\272\206\345\270\202\344\270\255\350\200\203\347\211\251\347\220\206\350\257\225\351\242\230\357\274\210A\345\215\267\357\274\211\357\274\210\345\216\237\345\215\267\347\211\210\357\274\211.pdf"
diff --git a/example_uploads/test2.jpg b/example_uploads/exams/test2.jpg
similarity index 100%
rename from example_uploads/test2.jpg
rename to example_uploads/exams/test2.jpg
diff --git a/example_uploads/test3.jpg b/example_uploads/exams/test3.jpg
similarity index 100%
rename from example_uploads/test3.jpg
rename to example_uploads/exams/test3.jpg
diff --git a/example_uploads/test4.pdf b/example_uploads/exams/test4.pdf
similarity index 100%
rename from example_uploads/test4.pdf
rename to example_uploads/exams/test4.pdf
diff --git a/example_uploads/test5.jpg b/example_uploads/exams/test5.jpg
similarity index 100%
rename from example_uploads/test5.jpg
rename to example_uploads/exams/test5.jpg
diff --git a/example_uploads/test6.jpg b/example_uploads/exams/test6.jpg
similarity index 100%
rename from example_uploads/test6.jpg
rename to example_uploads/exams/test6.jpg
diff --git "a/example_uploads/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\270\200\351\235\242.jpg" "b/example_uploads/exams/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\270\200\351\235\242.jpg"
similarity index 100%
rename from "example_uploads/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\270\200\351\235\242.jpg"
rename to "example_uploads/exams/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\270\200\351\235\242.jpg"
diff --git "a/example_uploads/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\272\214\351\235\242.jpg" "b/example_uploads/exams/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\272\214\351\235\242.jpg"
similarity index 100%
rename from "example_uploads/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\272\214\351\235\242.jpg"
rename to "example_uploads/exams/\345\260\217\345\255\246\346\225\260\345\255\246\350\257\225\345\215\267\346\211\253\346\217\217\347\254\254\344\272\214\351\235\242.jpg"
diff --git a/example_uploads/test.jpg b/example_uploads/notes/test.jpg
similarity index 100%
rename from example_uploads/test.jpg
rename to example_uploads/notes/test.jpg
diff --git a/frontend/LINEAR_DESIGN.md b/frontend/LINEAR_DESIGN.md
new file mode 100644
index 00000000..aacf6cc3
--- /dev/null
+++ b/frontend/LINEAR_DESIGN.md
@@ -0,0 +1,111 @@
+# Design System Inspiration of Linear
+
+## 1. Visual Theme & Atmosphere
+
+Linear's website is a masterclass in dark-mode-first product design — a near-black canvas (`#08090a`) where content emerges from darkness like starlight. The overall impression is one of extreme precision engineering: every element exists in a carefully calibrated hierarchy of luminance, from barely-visible borders (`rgba(255,255,255,0.05)`) to soft, luminous text (`#f7f8f8`). This is not a dark theme applied to a light design — it is darkness as the native medium, where information density is managed through subtle gradations of white opacity rather than color variation.
+
+The typography system is built entirely on Inter Variable with OpenType features `"cv01"` and `"ss03"` enabled globally, giving the typeface a cleaner, more geometric character. Inter is used at a remarkable range of weights — from 300 (light body) through 510 (medium, Linear's signature weight) to 590 (semibold emphasis). The 510 weight is particularly distinctive: it sits between regular and medium, creating a subtle emphasis that doesn't shout. At display sizes (72px, 64px, 48px), Inter uses aggressive negative letter-spacing (-1.584px to -1.056px), creating compressed, authoritative headlines that feel engineered rather than designed. Berkeley Mono serves as the monospace companion for code and technical labels, with fallbacks to ui-monospace, SF Mono, and Menlo.
+
+The color system is almost entirely achromatic — dark backgrounds with white/gray text — punctuated by a single brand accent: Linear's signature indigo-violet (`#5e6ad2` for backgrounds, `#7170ff` for interactive accents). This accent color is used sparingly and intentionally, appearing only on CTAs, active states, and brand elements. The border system uses ultra-thin, semi-transparent white borders (`rgba(255,255,255,0.05)` to `rgba(255,255,255,0.08)`) that create structure without visual noise, like wireframes drawn in moonlight.
+
+**Key Characteristics:**
+- Dark-mode-native: `#08090a` marketing background, `#0f1011` panel background, `#191a1b` elevated surfaces
+- Inter Variable with `"cv01", "ss03"` globally — geometric alternates for a cleaner aesthetic
+- Signature weight 510 (between regular and medium) for most UI text
+- Aggressive negative letter-spacing at display sizes (-1.584px at 72px, -1.056px at 48px)
+- Brand indigo-violet: `#5e6ad2` (bg) / `#7170ff` (accent) / `#828fff` (hover) — the only chromatic color in the system
+- Semi-transparent white borders throughout: `rgba(255,255,255,0.05)` to `rgba(255,255,255,0.08)`
+- Button backgrounds at near-zero opacity: `rgba(255,255,255,0.02)` to `rgba(255,255,255,0.05)`
+- Multi-layered shadows with inset variants for depth on dark surfaces
+- Success green (`#27a644`, `#10b981`) used only for status indicators
+
+## 2. Color Palette & Roles
+
+### Background Surfaces
+- **Marketing Black** (`#010102` / `#08090a`): The deepest background — the canvas for hero sections and marketing pages.
+- **Panel Dark** (`#0f1011`): Sidebar and panel backgrounds. One step up from the marketing black.
+- **Level 3 Surface** (`#191a1b`): Elevated surface areas, card backgrounds, dropdowns.
+- **Secondary Surface** (`#28282c`): The lightest dark surface — used for hover states and slightly elevated components.
+
+### Text & Content
+- **Primary Text** (`#f7f8f8`): Near-white with a barely-warm cast. Not pure white, preventing eye strain.
+- **Secondary Text** (`#d0d6e0`): Cool silver-gray for body text, descriptions.
+- **Tertiary Text** (`#8a8f98`): Muted gray for placeholders, metadata.
+- **Quaternary Text** (`#62666d`): The most subdued text — timestamps, disabled states.
+
+### Brand & Accent
+- **Brand Indigo** (`#5e6ad2`): Primary brand color — CTA button backgrounds, brand marks.
+- **Accent Violet** (`#7170ff`): Brighter variant for interactive elements — links, active states.
+- **Accent Hover** (`#828fff`): Lighter variant for hover states on accent elements.
+
+### Border & Divider
+- **Border Subtle** (`rgba(255,255,255,0.05)`): Ultra-subtle semi-transparent border — the default.
+- **Border Standard** (`rgba(255,255,255,0.08)`): Standard semi-transparent border for cards, inputs.
+- **Border Primary** (`#23252a`): Solid dark border for prominent separations.
+
+## 3. Typography Rules
+
+### Font Family
+- **Primary**: `Inter Variable` with `"cv01", "ss03"` OpenType features
+- **Monospace**: `Berkeley Mono`
+
+### Key Weights
+- **400**: Reading text
+- **510**: Emphasis/UI (Linear's signature weight)
+- **590**: Strong emphasis
+
+### Display Sizes Letter Spacing
+- 72px: -1.584px
+- 64px: -1.408px
+- 48px: -1.056px
+- 32px: -0.704px
+- Below 16px: normal
+
+## 4. Component Stylings
+
+### Buttons
+- **Ghost**: `rgba(255,255,255,0.02)` bg, `1px solid rgb(36,40,44)` border, 6px radius
+- **Subtle**: `rgba(255,255,255,0.04)` bg, 6px radius
+- **Primary Brand**: `#5e6ad2` bg, white text, 6px radius
+- **Pill**: transparent bg, 9999px radius, `1px solid rgb(35,37,42)` border
+
+### Cards & Containers
+- Background: `rgba(255,255,255,0.02)` to `rgba(255,255,255,0.05)` (never solid)
+- Border: `1px solid rgba(255,255,255,0.08)`
+- Radius: 8px (standard), 12px (featured)
+
+### Inputs
+- Background: `rgba(255,255,255,0.02)`
+- Border: `1px solid rgba(255,255,255,0.08)`
+- Radius: 6px
+
+## 5. Layout Principles
+
+### Spacing System
+- Base unit: 8px
+- Primary rhythm: 8px, 16px, 24px, 32px
+
+### Border Radius Scale
+- 2px: Inline badges, toolbar buttons
+- 6px: Buttons, inputs
+- 8px: Cards, dropdowns
+- 12px: Panels, featured cards
+- 9999px: Chips, pills
+
+## 6. Depth & Elevation
+
+Surface elevation via background opacity: `rgba(255,255,255, 0.02 -> 0.04 -> 0.05)` — never solid backgrounds on dark.
+
+## 7. Do's and Don'ts
+
+### Do
+- Use `#f7f8f8` for primary text — not pure `#ffffff`
+- Keep button backgrounds nearly transparent
+- Reserve brand indigo for primary CTAs only
+- Use semi-transparent white borders
+
+### Don't
+- Don't use pure white as primary text
+- Don't use solid colored backgrounds for buttons
+- Don't use weight 700 (bold) — max is 590
+- Don't use drop shadows for elevation on dark surfaces
diff --git a/frontend/app.html b/frontend/app.html
index 39a59650..8dfc7237 100644
--- a/frontend/app.html
+++ b/frontend/app.html
@@ -2,7 +2,7 @@
 <html lang="zh-CN">
   <head>
     <meta charset="UTF-8" />
-    <link rel="icon" type="image/svg+xml" href="/logo.svg" />
+    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <title>智卷错题本系统</title>
 
diff --git a/frontend/package.json b/frontend/package.json
index 19480577..27775cfb 100644
--- a/frontend/package.json
+++ b/frontend/package.json
@@ -12,6 +12,7 @@
   },
   "dependencies": {
     "@headlessui/vue": "^1.7.23",
+    "@tailwindcss/typography": "^0.5.19",
     "dompurify": "^3.3.2",
     "lucide-vue-next": "^0.577.0",
     "vue": "^3.5.24",
diff --git a/frontend/public/favicon.svg b/frontend/public/favicon.svg
new file mode 100644
index 00000000..b98ffb13
--- /dev/null
+++ b/frontend/public/favicon.svg
@@ -0,0 +1,28 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64">
+  <defs>
+    <linearGradient id="bg" x1="0" y1="0" x2="0" y2="1">
+      <stop offset="0%" stop-color="rgba(129,115,223,0.95)"/>
+      <stop offset="100%" stop-color="rgba(99,87,199,0.95)"/>
+    </linearGradient>
+    <pattern id="grid" width="8" height="8" patternUnits="userSpaceOnUse">
+      <path d="M 8 0 L 0 0 0 8" fill="none" stroke="rgba(255,255,255,0.08)" stroke-width="0.5"/>
+    </pattern>
+    <radialGradient id="mask-grad" cx="50%" cy="50%" r="50%">
+      <stop offset="30%" stop-color="white"/>
+      <stop offset="80%" stop-color="white" stop-opacity="0"/>
+    </radialGradient>
+    <mask id="grid-mask">
+      <rect width="64" height="64" fill="url(#mask-grad)"/>
+    </mask>
+  </defs>
+  <!-- 圆角背景 -->
+  <rect width="64" height="64" rx="14" fill="url(#bg)"/>
+  <!-- 顶部高光 -->
+  <rect x="0" y="0" width="64" height="1" rx="14" fill="rgba(255,255,255,0.15)"/>
+  <!-- 网格纹理 -->
+  <rect width="64" height="64" rx="14" fill="url(#grid)" mask="url(#grid-mask)"/>
+  <!-- 书本图标（缩放居中） -->
+  <g transform="translate(12, 14) scale(0.07)">
+    <path fill="#ffffff" d="M249.6 471.5c10.8 3.8 22.4-4.1 22.4-15.5l0-377.4c0-4.2-1.6-8.4-5-11C247.4 52 202.4 32 144 32C93.5 32 46.3 45.3 18.1 56.1C6.8 60.5 0 71.7 0 83.8L0 454.1c0 11.9 12.8 20.2 24.1 16.5C55.6 460.1 105.5 448 144 448c33.9 0 79 14 105.6 23.5zm76.8 0C353 462 398.1 448 432 448c38.5 0 88.4 12.1 119.9 22.6c11.3 3.8 24.1-4.6 24.1-16.5l0-370.3c0-12.1-6.8-23.3-18.1-27.6C529.7 45.3 482.5 32 432 32c-58.4 0-103.4 20-123 35.6c-3.3 2.6-5 6.8-5 11L304 456c0 11.4 11.7 19.3 22.4 15.5z"/>
+  </g>
+</svg>
diff --git a/frontend/public/logo.svg b/frontend/public/logo.svg
index 4dd7fb07..352aab88 100644
--- a/frontend/public/logo.svg
+++ b/frontend/public/logo.svg
@@ -1,7 +1,4 @@
-<svg viewBox="0 0 1024 1024" xmlns="http://www.w3.org/2000/svg" width="200" height="200">
-  <style>
-    path { fill: #2563EB; }
-    @media (prefers-color-scheme: dark) { path { fill: #ffffff; } }
-  </style>
-  <path d="M832 64c35.328 0 65.536 12.48 90.496 37.504 24.96 24.96 37.504 55.168 37.504 90.496v682.624c0 35.392-12.48 65.536-37.504 90.56-24.96 24.96-55.168 37.44-90.496 37.44H277.312c-31.232 0-58.752-10.048-82.56-30.208A124.288 124.288 0 0 1 151.04 896h51.584c26.048 0 48.384-9.088 67.072-27.328 18.624-18.176 28.288-40.32 28.864-66.432 0.64-26.048-7.936-48.64-25.728-67.712a92.736 92.736 0 0 0-65.728-30.464L149.312 704V629.376h53.376c25.6 0 47.68-8.896 66.24-26.624 18.56-17.664 28.416-39.296 29.632-64.896 0-25.6-8.704-52.224-26.432-70.784a92.8 92.8 0 0 0-64.96-29.632l-57.856-0.128V362.688h53.376c25.6 0 47.68-8.832 66.24-26.496 18.56-17.664 28.416-39.36 29.632-64.96 0-25.6-8.704-52.224-26.432-70.784a92.48 92.48 0 0 0-64.96-29.632l-56.064-0.128c5.248-30.72 19.776-56.192 43.584-76.352A124.16 124.16 0 0 1 277.312 64H832zM202.688 768c21.312 0 32 10.624 32 32 0 21.376-10.688 32-32 32H96c-21.312 0-32-10.624-32-32 0-21.376 10.688-32 32-32h106.688z m605.824-445.888c-30.144-30.144-60.352-30.144-90.496 0l-306.56 306.688a41.6 41.6 0 0 0-12.16 24.448l-8.64 63.808c-2.56 18.816 5.568 26.816 24.32 23.936l62.848-9.472a41.472 41.472 0 0 0 23.872-11.968l306.816-306.88c30.144-30.144 30.208-60.352 0-90.56z m-605.824 179.2c21.312 0 32 10.688 32 32 0 21.312-10.688 32-32 32H96c-21.312 0-32-10.688-32-32 0-21.312 10.688-32 32-32h106.688z m0-266.688c21.312 0 32 10.688 32 32 0 21.376-10.688 32-32 32H96c-21.312 0-32-10.624-32-32 0-21.312 10.688-32 32-32h106.688z"/>
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 576 512">
+  <!--! Font Awesome Free 6.7.2 by @fontawesome - https://fontawesome.com License - https://fontawesome.com/license/free -->
+  <path fill="#ffffff" d="M249.6 471.5c10.8 3.8 22.4-4.1 22.4-15.5l0-377.4c0-4.2-1.6-8.4-5-11C247.4 52 202.4 32 144 32C93.5 32 46.3 45.3 18.1 56.1C6.8 60.5 0 71.7 0 83.8L0 454.1c0 11.9 12.8 20.2 24.1 16.5C55.6 460.1 105.5 448 144 448c33.9 0 79 14 105.6 23.5zm76.8 0C353 462 398.1 448 432 448c38.5 0 88.4 12.1 119.9 22.6c11.3 3.8 24.1-4.6 24.1-16.5l0-370.3c0-12.1-6.8-23.3-18.1-27.6C529.7 45.3 482.5 32 432 32c-58.4 0-103.4 20-123 35.6c-3.3 2.6-5 6.8-5 11L304 456c0 11.4 11.7 19.3 22.4 15.5z"/>
 </svg>
diff --git a/frontend/src/App.vue b/frontend/src/App.vue
index 797b19c4..42ca3f26 100644
--- a/frontend/src/App.vue
+++ b/frontend/src/App.vue
@@ -1,21 +1,29 @@
 <script setup>
 import { onMounted } from 'vue'
-import AppLoading from './components/AppLoading.vue'
-import { usePageTransition } from './composables/usePageTransition.js'
 
+// ── 全局组件 ──────────────────────────────────────────
+import BaseLoading from '@/components/base/BaseLoading.vue'
+
+// ── 全局状态与逻辑 (Composables) ──────────────────────
+import { usePageTransition } from '@/composables/usePageTransition.js'
+import { useTheme } from '@/composables/useTheme.js'
+
+// 获取页面切换的加载状态及动画完成回调
 const { loading, notifyEnterCompleted } = usePageTransition()
+// 获取主题初始化方法
+const { initTheme } = useTheme()
 
-// 尽早应用已保存的主题，减少闪烁
+// ── 生命周期 ──────────────────────────────────────────
 onMounted(() => {
-  const saved = localStorage.getItem('theme')
-  const prefersDark = window.matchMedia?.('(prefers-color-scheme: dark)').matches
-  const initial = saved || (prefersDark ? 'dark' : 'light')
-  if (initial === 'dark') document.documentElement.classList.add('dark')
-  else document.documentElement.classList.remove('dark')
+  // 应用启动时初始化：根据本地存储或系统偏好设置深色/浅色模式
+  initTheme()
 })
 </script>
 
 <template>
+  <!-- 路由出口：渲染当前路径匹配的页面组件 -->
   <RouterView />
-  <AppLoading :visible="loading" @after-enter="notifyEnterCompleted" />
+
+  <!-- 全局加载遮罩：配合 usePageTransition 控制路由切换时的平滑过渡动画 -->
+  <BaseLoading :visible="loading" @after-enter="notifyEnterCompleted" />
 </template>
diff --git a/frontend/src/api.js b/frontend/src/api.js
index 9ebf1f6b..6924dd7b 100644
--- a/frontend/src/api.js
+++ b/frontend/src/api.js
@@ -75,11 +75,27 @@ export function uploadFiles(formData, { onProgress, onSuccess, onError, onAbort
   return xhr
 }
 
-export async function splitQuestions(modelProvider, modelName, { erase = false } = {}) {
+export async function runErase() {
+  const resp = await fetch('/api/erase', { method: 'POST' })
+  if (!resp.ok) throw new Error(`HTTP ${resp.status}`)
+  const data = await resp.json()
+  if (data && data.success) return data
+  throw new Error((data && data.error) || '擦除失败')
+}
+
+export async function runOcr() {
+  const resp = await fetch('/api/ocr', { method: 'POST' })
+  if (!resp.ok) throw new Error(`HTTP ${resp.status}`)
+  const data = await resp.json()
+  if (data && data.success) return data
+  throw new Error((data && data.error) || 'OCR 执行失败')
+}
+
+export async function splitQuestions(modelProvider, modelName) {
   const resp = await fetch('/api/split', {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },
-    body: JSON.stringify(_buildModelBody(modelProvider, modelName, { erase })),
+    body: JSON.stringify(_buildModelBody(modelProvider, modelName)),
   })
   if (!resp.ok) throw new Error(`HTTP ${resp.status}`)
   const data = await resp.json()
@@ -292,12 +308,143 @@ export async function fetchMessages(sessionId, { limit = 30, beforeId } = {}) {
   throw new Error((data && data.error) || '获取消息失败')
 }
 
-export async function streamChat(sessionId, message, modelProvider = 'openai', signal, modelName) {
+export async function streamChat(sessionId, message, modelProvider = 'openai', signal, modelName, { deepThink = false } = {}) {
   const opts = {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },
-    body: JSON.stringify(_buildModelBody(modelProvider, modelName, { message })),
+    body: JSON.stringify(_buildModelBody(modelProvider, modelName, { message, deep_think: deepThink })),
   }
   if (signal) opts.signal = signal
   return fetch(`/api/chat/${sessionId}/stream`, opts)
 }
+
+// ── 笔记模块 ─────────────────────────────────────────────
+
+/**
+ * 上传笔记图片 → OCR → LLM 整理 → 保存
+ * @param {FormData} formData - 包含 files、model_provider、model_name
+ * @param {Object} callbacks - { onProgress, onSuccess, onError }
+ * @returns {XMLHttpRequest} xhr 对象，可用于取消
+ */
+export function createNote(formData, { onProgress, onSuccess, onError } = {}) {
+  const xhr = new XMLHttpRequest()
+  xhr.open('POST', '/api/notes/')
+  xhr.withCredentials = true
+
+  if (onProgress) {
+    xhr.upload.onprogress = (e) => {
+      if (e.lengthComputable) onProgress(e.loaded / e.total)
+    }
+  }
+  xhr.onload = () => {
+    try {
+      const data = JSON.parse(xhr.responseText)
+      if (xhr.status >= 200 && xhr.status < 300 && data.success) {
+        onSuccess?.(data)
+      } else {
+        onError?.(data.error || '笔记创建失败')
+      }
+    } catch {
+      onError?.('解析响应失败')
+    }
+  }
+  xhr.onerror = () => onError?.('网络错误')
+  xhr.send(formData)
+  return xhr
+}
+
+/**
+ * 分页查询笔记列表
+ */
+export async function fetchNotes(params = {}) {
+  const query = new URLSearchParams()
+  if (params.page) query.set('page', params.page)
+  if (params.limit) query.set('limit', params.limit)
+  if (params.subject) query.set('subject', params.subject)
+  if (params.knowledge_tag) query.set('knowledge_tag', params.knowledge_tag)
+  if (params.keyword) query.set('keyword', params.keyword)
+
+  const resp = await fetch(`/api/notes/?${query}`)
+  const data = await resp.json()
+  if (data.success) return data
+  throw new Error(data.error || '获取笔记列表失败')
+}
+
+/**
+ * 获取单条笔记详情
+ */
+export async function fetchNote(noteId) {
+  const resp = await fetch(`/api/notes/${noteId}`)
+  const data = await resp.json()
+  if (data.success) return data.note
+  throw new Error(data.error || '获取笔记详情失败')
+}
+
+/**
+ * 更新笔记
+ */
+export async function updateNote(noteId, payload) {
+  const resp = await fetch(`/api/notes/${noteId}`, {
+    method: 'PATCH',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify(payload),
+  })
+  const data = await resp.json()
+  if (data.success) return data.note
+  throw new Error(data.error || '更新笔记失败')
+}
+
+/**
+ * 删除笔记
+ */
+export async function deleteNote(noteId) {
+  const resp = await fetch(`/api/notes/${noteId}`, { method: 'DELETE' })
+  const data = await resp.json()
+  if (data.success) return true
+  throw new Error(data.error || '删除笔记失败')
+}
+
+// ── 独立对话 ─────────────────────────────────────────────
+
+/** 创建独立对话 */
+export async function createIndependentChat(title = '新对话') {
+  const resp = await fetch('/api/chat', {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ title }),
+  })
+  const data = await resp.json()
+  if (data.success) return data.session
+  throw new Error(data.error || '创建对话失败')
+}
+
+/** 获取用户的独立对话列表 */
+export async function fetchMyChatSessions(params = {}) {
+  const query = new URLSearchParams()
+  if (params.page) query.set('page', params.page)
+  if (params.limit) query.set('limit', params.limit)
+  const resp = await fetch(`/api/chat/my-sessions?${query}`)
+  const data = await resp.json()
+  if (data.success) return data
+  throw new Error(data.error || '获取对话列表失败')
+}
+
+/** 更新对话标题 */
+export async function updateChatTitle(sessionId, title) {
+  const resp = await fetch(`/api/chat/${sessionId}`, {
+    method: 'PATCH',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ title }),
+  })
+  const data = await resp.json()
+  if (data.success) return data.session
+  throw new Error(data.error || '更新标题失败')
+}
+
+/** 删除对话 */
+export async function deleteChat(sessionId) {
+  const resp = await fetch(`/api/chat/${sessionId}`, { method: 'DELETE' })
+  const data = await resp.json()
+  if (data.success) return true
+  throw new Error(data.error || '删除对话失败')
+}
diff --git a/frontend/src/assets/vue.svg b/frontend/src/assets/vue.svg
deleted file mode 100644
index 770e9d33..00000000
--- a/frontend/src/assets/vue.svg
+++ /dev/null
@@ -1 +0,0 @@
-<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" class="iconify iconify--logos" width="37.07" height="36" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 198"><path fill="#41B883" d="M204.8 0H256L128 220.8L0 0h97.92L128 51.2L157.44 0h47.36Z"></path><path fill="#41B883" d="m0 0l128 220.8L256 0h-51.2L128 132.48L50.56 0H0Z"></path><path fill="#35495E" d="M50.56 0L128 133.12L204.8 0h-47.36L128 51.2L97.92 0H50.56Z"></path></svg>
\ No newline at end of file
diff --git a/frontend/src/components/AuthLayout.vue b/frontend/src/components/AuthLayout.vue
deleted file mode 100644
index c0045e43..00000000
--- a/frontend/src/components/AuthLayout.vue
+++ /dev/null
@@ -1,177 +0,0 @@
-<script setup>
-import { ref, onMounted } from 'vue'
-import { useRoute, useRouter } from 'vue-router'
-
-const route = useRoute()
-const router = useRouter()
-const transitionName = ref('auth-slide-left')
-
-onMounted(() => {
-  document.documentElement.classList.add('dark')
-  localStorage.setItem('theme', 'dark')
-})
-
-router.beforeEach((to, from) => {
-  const toOrder = to.meta.order ?? 0
-  const fromOrder = from.meta.order ?? 0
-  transitionName.value = toOrder >= fromOrder ? 'auth-slide-left' : 'auth-slide-right'
-})
-
-const FEATURES = [
-  { icon: 'fa-camera', text: '拍照上传，AI 自动识别题目与公式' },
-  { icon: 'fa-brain', text: 'LangGraph Agent 智能拆题纠错' },
-  { icon: 'fa-tags', text: '知识点自动打标，构建个人图谱' },
-  { icon: 'fa-file-export', text: '一键导出 Markdown / PDF 错题本' },
-]
-</script>
-
-<template>
-  <div class="min-h-screen flex bg-slate-50 dark:bg-[#0A0A0F]">
-
-
-    <!-- 左侧品牌区（大屏显示） -->
-    <div class="hidden lg:flex flex-col justify-between w-[52%] relative overflow-hidden bg-gradient-to-br from-blue-600 to-indigo-700 dark:from-indigo-950 dark:to-[#0A0A0F] p-12">
-      <!-- 背景装饰光晕 -->
-      <div class="absolute inset-0 pointer-events-none">
-        <div class="absolute top-[-10%] left-[-10%] w-[500px] h-[500px] rounded-full bg-blue-400/20 dark:bg-indigo-500/10 blur-[120px]"></div>
-        <div class="absolute bottom-[-10%] right-[-10%] w-[400px] h-[400px] rounded-full bg-indigo-400/20 dark:bg-blue-500/10 blur-[100px]"></div>
-        <!-- 网格纹理 -->
-        <div class="absolute inset-0 opacity-[0.04]" style="background-image: linear-gradient(rgba(255,255,255,0.5) 1px, transparent 1px), linear-gradient(90deg, rgba(255,255,255,0.5) 1px, transparent 1px); background-size: 40px 40px;"></div>
-      </div>
-
-      <!-- 顶部 Logo -->
-      <div class="relative flex items-center gap-3">
-        <div class="relative group">
-          <div class="absolute inset-0 bg-blue-500/60 dark:bg-indigo-500/60 blur-xl opacity-0 group-hover:opacity-100 transition duration-500 rounded-2xl"></div>
-          <div class="relative bg-gradient-to-br from-blue-500 to-indigo-700 dark:from-indigo-500 dark:to-indigo-800 p-2.5 rounded-2xl shadow-md shadow-indigo-500/30 border border-white/10">
-            <img src="/logo.svg" class="w-7 h-7 brightness-0 invert" alt="logo" />
-          </div>
-        </div>
-        <span class="text-2xl font-bold bg-clip-text text-transparent bg-gradient-to-r from-white via-indigo-200 to-indigo-200 tracking-wide">智卷错题本</span>
-      </div>
-
-      <!-- 中部主文案 -->
-      <div class="relative">
-        <div class="inline-flex items-center gap-2 px-3 py-1 rounded-full bg-white/10 border border-white/20 text-xs font-bold text-blue-100 dark:text-indigo-300 tracking-widest uppercase mb-6">
-          <span class="w-1.5 h-1.5 rounded-full bg-emerald-400 animate-pulse"></span>
-          AI 驱动的错题管理
-        </div>
-        <h2 class="text-4xl font-extrabold text-white leading-tight mb-4">
-          重塑错题整理<br />
-          <span class="text-blue-200 dark:text-indigo-300">一键生成知识图谱</span>
-        </h2>
-        <p class="text-base text-blue-100/80 dark:text-slate-400 leading-relaxed max-w-sm">
-          专为中学生与大学生研发。上传凌乱试卷，AI 自动完成图片分割、OCR 纠错及 LaTeX 公式还原。
-        </p>
-
-        <!-- 特性列表 -->
-        <ul class="mt-8 space-y-4">
-          <li v-for="f in FEATURES" :key="f.text" class="flex items-center gap-3 text-sm text-blue-100/90 dark:text-slate-300">
-            <div class="flex h-8 w-8 shrink-0 items-center justify-center rounded-xl bg-white/10 border border-white/15">
-              <i :class="`fas ${f.icon} text-xs text-white`"></i>
-            </div>
-            {{ f.text }}
-          </li>
-        </ul>
-      </div>
-
-      <!-- 底部版权 -->
-      <div class="relative text-xs text-blue-200/50 dark:text-slate-600">
-        © {{ new Date().getFullYear() }} 智卷错题本 · All rights reserved
-      </div>
-    </div>
-
-    <!-- 右侧表单区 -->
-    <div class="flex-1 flex flex-col items-center justify-center px-4 py-12 lg:px-16 relative">
-
-      <!-- 返回主页 -->
-      <a href="/" class="absolute top-6 right-6 inline-flex items-center gap-2 px-4 py-2 rounded-full text-sm font-semibold text-slate-700 dark:text-slate-200 bg-white/80 dark:bg-white/15 backdrop-blur-md border border-slate-200/60 dark:border-white/10 hover:bg-white dark:hover:bg-white/25 transition-all shadow-sm">
-        <i class="fas fa-arrow-left text-xs"></i>
-        返回主页
-      </a>
-
-      <!-- 移动端 Logo（小屏显示） -->
-      <div class="lg:hidden text-center mb-8">
-        <div class="relative inline-flex mb-4 group">
-          <div class="absolute inset-0 bg-blue-500/60 dark:bg-indigo-500/60 blur-xl opacity-0 group-hover:opacity-100 transition duration-500 rounded-2xl"></div>
-          <div class="relative bg-gradient-to-br from-blue-500 to-indigo-700 dark:from-indigo-500 dark:to-indigo-800 p-2.5 rounded-2xl shadow-md shadow-indigo-500/30 border border-white/10">
-            <img src="/logo.svg" class="w-7 h-7 brightness-0 invert" alt="logo" />
-          </div>
-        </div>
-        <h1 class="text-2xl font-bold bg-clip-text text-transparent bg-gradient-to-r from-blue-700 to-indigo-700 dark:from-white dark:via-indigo-200 dark:to-indigo-200 tracking-wide">智卷错题本</h1>
-      </div>
-
-      <div class="w-full max-w-sm">
-        <!-- 标题 -->
-        <div class="mb-8">
-          <h3 class="text-2xl font-bold text-slate-900 dark:text-white">
-            {{ route.path === '/auth/login' ? '欢迎回来' : '创建账户' }}
-          </h3>
-          <p class="text-sm text-slate-500 dark:text-slate-400 mt-1">
-            {{ route.path === '/auth/login' ? '登录以继续使用你的错题本' : '免费注册，开始智能错题整理' }}
-          </p>
-        </div>
-
-        <!-- Tab 切换 -->
-        <div class="flex rounded-xl bg-slate-100/80 dark:bg-white/5 border border-slate-200/60 dark:border-white/10 p-1 mb-6">
-          <RouterLink
-            to="/auth/login"
-            class="flex-1 py-2 text-sm font-bold rounded-xl text-center transition-all"
-            :class="route.path === '/auth/login'
-              ? 'bg-white dark:bg-white/10 text-blue-600 dark:text-blue-400 shadow-sm'
-              : 'text-slate-500 dark:text-slate-400 hover:text-slate-700 dark:hover:text-slate-200'"
-          >登录</RouterLink>
-          <RouterLink
-            to="/auth/register"
-            class="flex-1 py-2 text-sm font-bold rounded-xl text-center transition-all"
-            :class="route.path === '/auth/register'
-              ? 'bg-white dark:bg-white/10 text-blue-600 dark:text-blue-400 shadow-sm'
-              : 'text-slate-500 dark:text-slate-400 hover:text-slate-700 dark:hover:text-slate-200'"
-          >注册</RouterLink>
-        </div>
-
-        <!-- 表单内容（滑动过渡） -->
-        <RouterView v-slot="{ Component }">
-          <Transition :name="transitionName" mode="out-in">
-            <component :is="Component" :key="route.path" />
-          </Transition>
-        </RouterView>
-
-        <!-- 底部版权（移动端） -->
-        <p class="lg:hidden text-center text-xs text-slate-400 dark:text-slate-600 mt-8">
-          © {{ new Date().getFullYear() }} 智卷错题本
-        </p>
-      </div>
-    </div>
-
-  </div>
-</template>
-
-<style>
-/* 禁用浏览器自动填充背景色 */
-input:-webkit-autofill,
-input:-webkit-autofill:hover,
-input:-webkit-autofill:focus {
-  transition: background-color 9999s ease;
-  -webkit-text-fill-color: #f1f5f9;
-  caret-color: #f1f5f9;
-}
-</style>
-
-<style scoped>
-/* 向左滑（登录 → 注册） */
-.auth-slide-left-enter-active,
-.auth-slide-left-leave-active {
-  transition: opacity 0.22s ease, transform 0.22s ease;
-}
-.auth-slide-left-enter-from  { opacity: 0; transform: translateX(32px); }
-.auth-slide-left-leave-to   { opacity: 0; transform: translateX(-32px); }
-
-/* 向右滑（注册 → 登录） */
-.auth-slide-right-enter-active,
-.auth-slide-right-leave-active {
-  transition: opacity 0.22s ease, transform 0.22s ease;
-}
-.auth-slide-right-enter-from { opacity: 0; transform: translateX(-32px); }
-.auth-slide-right-leave-to  { opacity: 0; transform: translateX(32px); }
-</style>
diff --git a/frontend/src/components/AuthView.vue b/frontend/src/components/AuthView.vue
deleted file mode 100644
index be632995..00000000
--- a/frontend/src/components/AuthView.vue
+++ /dev/null
@@ -1,253 +0,0 @@
-<template>
-  <div class="min-h-screen flex items-center justify-center bg-gradient-to-br from-slate-50 to-slate-100 dark:from-[#0A0A0F] dark:to-slate-900 px-4">
-    <div class="w-full max-w-md">
-      <!-- Logo / Title -->
-      <div class="text-center mb-8">
-        <div class="relative inline-flex mb-4">
-          <div class="absolute inset-0 bg-blue-500 dark:bg-indigo-500 blur-lg opacity-40 dark:opacity-60 rounded-xl"></div>
-          <div class="relative bg-blue-600 dark:bg-gradient-to-br dark:from-indigo-500 dark:to-indigo-600 p-3 rounded-xl shadow-md">
-            <img src="/logo.svg" class="w-8 h-8 brightness-0 invert" alt="logo" />
-          </div>
-        </div>
-        <h1 class="text-2xl font-bold text-slate-800 dark:text-white">错题本</h1>
-        <p class="text-sm text-slate-500 dark:text-slate-400 mt-1">智能错题管理系统</p>
-      </div>
-
-      <!-- Card -->
-      <div class="bg-white/70 dark:bg-white/5 backdrop-blur-xl rounded-3xl border border-slate-200/60 dark:border-white/10 shadow-md p-8">
-        <!-- Tabs -->
-        <div class="flex rounded-xl bg-slate-100/80 dark:bg-white/5 p-1 mb-6">
-          <button
-            @click="activeTab = 'login'"
-            :class="[
-              'flex-1 py-2 text-sm font-bold rounded-xl transition-all',
-              activeTab === 'login'
-                ? 'bg-white dark:bg-white/10 text-blue-600 dark:text-blue-400 shadow-sm'
-                : 'text-slate-500 dark:text-slate-400 hover:text-slate-700 dark:hover:text-slate-200'
-            ]"
-          >
-            登录
-          </button>
-          <button
-            @click="activeTab = 'register'"
-            :class="[
-              'flex-1 py-2 text-sm font-bold rounded-xl transition-all',
-              activeTab === 'register'
-                ? 'bg-white dark:bg-white/10 text-blue-600 dark:text-blue-400 shadow-sm'
-                : 'text-slate-500 dark:text-slate-400 hover:text-slate-700 dark:hover:text-slate-200'
-            ]"
-          >
-            注册
-          </button>
-        </div>
-
-        <!-- Login Form -->
-        <form v-if="activeTab === 'login'" @submit.prevent="handleLogin" class="space-y-4">
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">邮箱或用户名</label>
-            <input
-              v-model="loginForm.email"
-              type="text"
-              required
-              autocomplete="username"
-              placeholder="邮箱 或 用户名"
-              class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-            />
-          </div>
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">密码</label>
-            <div class="relative">
-              <input
-                v-model="loginForm.password"
-                :type="showLoginPwd ? 'text' : 'password'"
-                required
-                autocomplete="current-password"
-                placeholder="请输入密码"
-                class="w-full h-10 px-4 pr-11 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-              />
-              <button
-                type="button"
-                @click="showLoginPwd = !showLoginPwd"
-                class="absolute right-3 top-1/2 -translate-y-1/2 text-slate-400 hover:text-slate-600 dark:hover:text-slate-200 transition-colors"
-              >
-                <i :class="showLoginPwd ? 'fas fa-eye-slash' : 'fas fa-eye'"></i>
-              </button>
-            </div>
-          </div>
-
-          <p v-if="loginError" class="text-sm text-rose-500 dark:text-rose-400 flex items-center gap-2">
-            <i class="fas fa-circle-exclamation"></i>{{ loginError }}
-          </p>
-
-          <button
-            type="submit"
-            :disabled="loading"
-            class="w-full h-12 bg-blue-600 hover:bg-blue-700 disabled:bg-blue-400 text-white font-bold rounded-xl transition-all shadow-md shadow-blue-500/30 hover:shadow-blue-500/50 disabled:cursor-not-allowed disabled:shadow-none flex items-center justify-center gap-2 text-sm"
-          >
-            <i v-if="loading" class="fas fa-spinner fa-spin"></i>
-            <span>{{ loading ? '登录中...' : '登录' }}</span>
-          </button>
-        </form>
-
-        <!-- Register Form -->
-        <form v-if="activeTab === 'register'" @submit.prevent="handleRegister" class="space-y-4">
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">用户名</label>
-            <input
-              v-model="registerForm.username"
-              type="text"
-              required
-              autocomplete="username"
-              placeholder="您的昵称"
-              maxlength="50"
-              class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-            />
-          </div>
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">邮箱</label>
-            <input
-              v-model="registerForm.email"
-              type="email"
-              required
-              autocomplete="email"
-              placeholder="your@email.com"
-              class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-            />
-          </div>
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">密码</label>
-            <div class="relative">
-              <input
-                v-model="registerForm.password"
-                :type="showRegPwd ? 'text' : 'password'"
-                required
-                autocomplete="new-password"
-                placeholder="至少 6 位"
-                minlength="6"
-                class="w-full h-10 px-4 pr-11 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-              />
-              <button
-                type="button"
-                @click="showRegPwd = !showRegPwd"
-                class="absolute right-3 top-1/2 -translate-y-1/2 text-slate-400 hover:text-slate-600 dark:hover:text-slate-200 transition-colors"
-              >
-                <i :class="showRegPwd ? 'fas fa-eye-slash' : 'fas fa-eye'"></i>
-              </button>
-            </div>
-          </div>
-          <div>
-            <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">确认密码</label>
-            <input
-              v-model="registerForm.confirm"
-              type="password"
-              required
-              autocomplete="new-password"
-              placeholder="再次输入密码"
-              class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-              :class="{ 'border-rose-400 dark:border-rose-500': registerForm.confirm && registerForm.password !== registerForm.confirm }"
-            />
-            <p v-if="registerForm.confirm && registerForm.password !== registerForm.confirm" class="text-xs text-rose-500 mt-1">
-              两次密码不一致
-            </p>
-          </div>
-
-          <p v-if="registerError" class="text-sm text-rose-500 dark:text-rose-400 flex items-center gap-2">
-            <i class="fas fa-circle-exclamation"></i>{{ registerError }}
-          </p>
-          <p v-if="registerSuccess" class="text-sm text-emerald-600 dark:text-emerald-400 flex items-center gap-2">
-            <i class="fas fa-circle-check"></i>{{ registerSuccess }}
-          </p>
-
-          <button
-            type="submit"
-            :disabled="loading || (registerForm.confirm && registerForm.password !== registerForm.confirm)"
-            class="w-full h-12 bg-blue-600 hover:bg-blue-700 disabled:bg-blue-400 text-white font-bold rounded-xl transition-all shadow-md shadow-blue-500/30 hover:shadow-blue-500/50 disabled:cursor-not-allowed disabled:shadow-none flex items-center justify-center gap-2 text-sm"
-          >
-            <i v-if="loading" class="fas fa-spinner fa-spin"></i>
-            <span>{{ loading ? '注册中...' : '创建账户' }}</span>
-          </button>
-        </form>
-      </div>
-
-      <p class="text-center text-xs text-slate-400 dark:text-slate-600 mt-6">
-        错题本智能管理系统 &copy; {{ new Date().getFullYear() }}
-      </p>
-    </div>
-  </div>
-</template>
-
-<script setup>
-import { ref, reactive } from 'vue'
-import { useRouter } from 'vue-router'
-import { useAuth } from '../composables/useAuth.js'
-
-const router = useRouter()
-const { currentUser } = useAuth()
-
-const activeTab = ref('login')
-const loading = ref(false)
-const showLoginPwd = ref(false)
-const showRegPwd = ref(false)
-const loginError = ref('')
-const registerError = ref('')
-const registerSuccess = ref('')
-
-const loginForm = reactive({ email: '', password: '' })
-const registerForm = reactive({ username: '', email: '', password: '', confirm: '' })
-
-async function handleLogin() {
-  loginError.value = ''
-  loading.value = true
-  try {
-    const res = await fetch('/api/auth/login', {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ email: loginForm.email, password: loginForm.password })
-    })
-    const data = await res.json()
-    if (!res.ok) {
-      loginError.value = data.error || '登录失败'
-    } else {
-      currentUser.value = data.user
-      router.push('/app')
-    }
-  } catch {
-    loginError.value = '网络错误，请重试'
-  } finally {
-    loading.value = false
-  }
-}
-
-async function handleRegister() {
-  registerError.value = ''
-  registerSuccess.value = ''
-  if (registerForm.password !== registerForm.confirm) {
-    registerError.value = '两次密码不一致'
-    return
-  }
-  loading.value = true
-  try {
-    const res = await fetch('/api/auth/register', {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({
-        username: registerForm.username,
-        email: registerForm.email,
-        password: registerForm.password
-      })
-    })
-    const data = await res.json()
-    if (!res.ok) {
-      registerError.value = data.error || '注册失败'
-    } else {
-      registerSuccess.value = '注册成功！正在登录...'
-      currentUser.value = data.user
-      router.push('/app')
-    }
-  } catch {
-    registerError.value = '网络错误，请重试'
-  } finally {
-    loading.value = false
-  }
-}
-</script>
diff --git a/frontend/src/components/CalendarPicker.vue b/frontend/src/components/CalendarPicker.vue
deleted file mode 100644
index ea159fe6..00000000
--- a/frontend/src/components/CalendarPicker.vue
+++ /dev/null
@@ -1,143 +0,0 @@
-<script setup>
-import { ref, computed } from 'vue'
-import { useClickOutside } from '../composables/useClickOutside.js'
-
-const props = defineProps({
-  modelValue: { type: String, default: '' },
-  label: { type: String, default: '选择日期' },
-  align: { type: String, default: 'left' },
-})
-
-const emit = defineEmits(['update:modelValue'])
-
-const open = ref(false)
-const calYear = ref(new Date().getFullYear())
-const calMonth = ref(new Date().getMonth())
-const WEEKDAYS = ['一', '二', '三', '四', '五', '六', '日']
-
-const toggle = () => {
-  if (open.value) { open.value = false; return }
-  const d = props.modelValue ? new Date(props.modelValue) : new Date()
-  calYear.value = d.getFullYear()
-  calMonth.value = d.getMonth()
-  open.value = true
-}
-
-const calDays = computed(() => {
-  const y = calYear.value
-  const m = calMonth.value
-  let dow = new Date(y, m, 1).getDay() - 1
-  if (dow < 0) dow = 6
-  const dim = new Date(y, m + 1, 0).getDate()
-  const prev = new Date(y, m, 0).getDate()
-  const days = []
-  for (let i = dow - 1; i >= 0; i--) days.push({ day: prev - i, cur: false })
-  for (let d = 1; d <= dim; d++) days.push({ day: d, cur: true })
-  while (days.length < 42) days.push({ day: days.length - dow - dim + 1, cur: false })
-  return days
-})
-
-const prevMonth = () => {
-  if (calMonth.value === 0) { calMonth.value = 11; calYear.value-- }
-  else calMonth.value--
-}
-
-const nextMonth = () => {
-  if (calMonth.value === 11) { calMonth.value = 0; calYear.value++ }
-  else calMonth.value++
-}
-
-const selectDate = (d) => {
-  if (!d.cur) return
-  const s = `${calYear.value}-${String(calMonth.value + 1).padStart(2, '0')}-${String(d.day).padStart(2, '0')}`
-  emit('update:modelValue', s)
-  open.value = false
-}
-
-const selectToday = () => {
-  const n = new Date()
-  calYear.value = n.getFullYear()
-  calMonth.value = n.getMonth()
-  selectDate({ day: n.getDate(), cur: true })
-}
-
-const clearDate = () => {
-  emit('update:modelValue', '')
-  open.value = false
-}
-
-const isDayToday = (d) => {
-  if (!d.cur) return false
-  const n = new Date()
-  return d.day === n.getDate() && calMonth.value === n.getMonth() && calYear.value === n.getFullYear()
-}
-
-const isDaySel = (d) => {
-  if (!d.cur || !props.modelValue) return false
-  const x = new Date(props.modelValue)
-  return d.day === x.getDate() && calMonth.value === x.getMonth() && calYear.value === x.getFullYear()
-}
-
-useClickOutside('.custom-cal-wrapper', () => { open.value = false })
-</script>
-
-<template>
-  <div class="custom-cal-wrapper relative w-full min-w-0">
-    <div class="group flex h-11 w-full cursor-pointer items-center rounded-xl border bg-white/60 px-4 text-sm font-bold backdrop-blur-xl transition-all hover:border-slate-300 hover:bg-white/80 hover:shadow-sm dark:border-white/10 dark:bg-white/[0.03] dark:hover:border-white/20 dark:hover:shadow-white/5"
-         :class="open ? 'border-white/30 bg-white/80 shadow-sm ring-2 ring-white/10 dark:border-white/20 dark:bg-white/[0.06] dark:ring-white/5' : 'border-slate-200/60'"
-         @click.stop="toggle">
-      <i class="fa-regular fa-calendar mr-2 text-sm text-slate-400 transition-colors group-hover:text-slate-500 dark:text-slate-500 dark:group-hover:text-slate-300"></i>
-      <span v-if="!modelValue" class="truncate text-sm font-bold text-slate-400 dark:text-slate-500">{{ label }}</span>
-      <span v-else class="truncate text-sm font-bold text-slate-800 dark:text-slate-200">{{ modelValue }}</span>
-    </div>
-    <Transition name="cal-dropdown">
-      <div v-if="open"
-           class="absolute top-full z-50 mt-2 min-w-full w-64 overflow-hidden rounded-2xl border border-slate-200/60 bg-white/95 p-4 shadow-xl backdrop-blur-3xl dark:border-white/10 dark:bg-[#12121A]/90 dark:bg-gradient-to-b dark:from-white/[0.08] dark:to-transparent dark:backdrop-blur-3xl dark:shadow-[0_20px_50px_rgba(0,0,0,0.6)]"
-           :class="align === 'right' ? 'right-0' : 'left-0'">
-        <!-- 月份导航 -->
-        <div class="mb-4 flex items-center justify-between">
-          <button @click.stop="prevMonth" class="flex h-8 w-8 items-center justify-center rounded-lg text-slate-400 hover:bg-slate-100 hover:text-slate-600 dark:hover:bg-white/5 dark:hover:text-white transition-colors">
-            <i class="fa-solid fa-chevron-left text-xs"></i>
-          </button>
-          <span class="text-sm font-black text-slate-700 dark:text-white">{{ calYear }}年{{ calMonth + 1 }}月</span>
-          <button @click.stop="nextMonth" class="flex h-8 w-8 items-center justify-center rounded-lg text-slate-400 hover:bg-slate-100 hover:text-slate-600 dark:hover:bg-white/5 dark:hover:text-white transition-colors">
-            <i class="fa-solid fa-chevron-right text-xs"></i>
-          </button>
-        </div>
-        <!-- 星期标题 -->
-        <div class="mb-2 grid grid-cols-7">
-          <span v-for="w in WEEKDAYS" :key="w" class="py-1 text-center text-xs font-black text-slate-400 dark:text-slate-500">{{ w }}</span>
-        </div>
-        <!-- 日期网格 -->
-        <div class="grid grid-cols-7 place-items-center gap-1">
-          <button v-for="(d, i) in calDays" :key="i" @click.stop="selectDate(d)"
-            class="flex h-8 w-8 items-center justify-center rounded-xl text-xs font-bold transition-all"
-            :class="[
-              !d.cur ? 'text-slate-300 dark:text-slate-600 opacity-30' : 'text-slate-700 dark:text-slate-200 cursor-pointer hover:bg-blue-50 dark:hover:bg-white/10',
-              isDaySel(d) ? '!bg-blue-600 !text-white shadow-md shadow-blue-500/20 dark:!bg-indigo-500' : '',
-              isDayToday(d) && !isDaySel(d) ? 'ring-1 ring-blue-400 text-blue-600 dark:text-indigo-400 dark:ring-indigo-500/50' : ''
-            ]">
-            {{ d.day }}
-          </button>
-        </div>
-        <!-- 底部操作 -->
-        <div class="mt-4 flex justify-between border-t border-slate-100 pt-3 dark:border-white/5">
-          <button @click.stop="clearDate" class="text-xs font-bold text-slate-400 hover:text-rose-500 dark:text-slate-500 dark:hover:text-rose-400 transition-colors">清除</button>
-          <button @click.stop="selectToday" class="text-xs font-bold text-blue-600 hover:text-blue-700 dark:text-indigo-400 dark:hover:text-indigo-300 transition-colors">今天</button>
-        </div>
-      </div>
-    </Transition>
-  </div>
-</template>
-
-<style scoped>
-.cal-dropdown-enter-active,
-.cal-dropdown-leave-active {
-  transition: all 0.2s cubic-bezier(0.16, 1, 0.3, 1);
-}
-.cal-dropdown-enter-from,
-.cal-dropdown-leave-to {
-  opacity: 0;
-  transform: translateY(-6px);
-}
-</style>
diff --git a/frontend/src/components/CustomSelect.vue b/frontend/src/components/CustomSelect.vue
deleted file mode 100644
index 3abd0aa4..00000000
--- a/frontend/src/components/CustomSelect.vue
+++ /dev/null
@@ -1,78 +0,0 @@
-<script setup>
-import { ref } from 'vue'
-import { useClickOutside } from '../composables/useClickOutside.js'
-
-const props = defineProps({
-  modelValue: { type: String, default: '' },
-  options: { type: Array, default: () => [] },
-  label: { type: String, default: '' },
-  placeholder: { type: String, default: '全部' },
-  icon: { type: Function, default: null },
-  widthClass: { type: String, default: '' },
-})
-
-const emit = defineEmits(['update:modelValue'])
-const open = ref(false)
-
-useClickOutside('.custom-select-wrapper', () => { open.value = false })
-
-const toggle = () => { open.value = !open.value }
-const select = (val) => { emit('update:modelValue', val); open.value = false }
-</script>
-
-<template>
-  <div class="custom-select-wrapper relative" :class="widthClass">
-    <label v-if="label" class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500 dark:text-slate-500">{{ label }}</label>
-    <button
-      type="button"
-      @click.stop="toggle"
-      class="flex h-11 w-full items-center justify-between rounded-xl border bg-white/60 px-4 text-left text-sm font-bold backdrop-blur-xl transition-all hover:border-slate-300 hover:bg-white/80 hover:shadow-sm dark:border-white/10 dark:bg-white/[0.03] dark:hover:border-white/20 dark:hover:shadow-white/5"
-      :class="[
-        open ? 'border-white/30 bg-white/80 shadow-sm ring-2 ring-white/10 dark:border-white/20 dark:bg-white/[0.06] dark:ring-white/5' : 'border-slate-200/60',
-        modelValue ? 'text-slate-800 dark:text-slate-200' : 'text-slate-400 dark:text-slate-500',
-      ]"
-    >
-      <span class="truncate">{{ modelValue || placeholder }}</span>
-      <i class="fa-solid fa-chevron-down ml-2 text-xs text-slate-400 transition-transform duration-300 dark:text-slate-500" :class="open ? 'rotate-180' : ''"></i>
-    </button>
-    <Transition name="dropdown">
-      <div v-if="open" class="absolute left-0 top-full z-50 mt-2 min-w-full w-max overflow-hidden rounded-2xl border border-slate-200/60 bg-white/95 p-1.5 shadow-xl backdrop-blur-3xl dark:border-white/10 dark:bg-[#12121A]/90 dark:bg-gradient-to-b dark:from-white/[0.08] dark:to-transparent dark:backdrop-blur-3xl dark:shadow-[0_20px_50px_rgba(0,0,0,0.6)]">
-        <div class="no-scrollbar max-h-56 space-y-0.5 overflow-y-auto">
-          <button
-            type="button"
-            @click.stop="select('')"
-            class="group flex w-full items-center gap-2 rounded-xl px-4 py-2.5 text-left text-sm font-bold transition-all hover:bg-white/60 hover:text-slate-800 dark:hover:bg-white/10 dark:hover:text-slate-100"
-            :class="modelValue === '' ? 'bg-white/60 text-slate-800 dark:bg-white/10 dark:text-slate-100' : 'text-slate-600 dark:text-slate-300'"
-          >
-            <i v-if="icon" class="fa-solid fa-layer-group text-xs opacity-70"></i>
-            <span>{{ placeholder }}</span>
-            <i v-if="modelValue === ''" class="fa-solid fa-check ml-auto text-[10px]"></i>
-          </button>
-          <button
-            v-for="opt in options" :key="opt"
-            type="button"
-            @click.stop="select(opt)"
-            class="group flex w-full items-center gap-2 rounded-xl px-4 py-2.5 text-left text-sm font-bold transition-all hover:bg-white/60 hover:text-slate-800 dark:hover:bg-white/10 dark:hover:text-slate-100"
-            :class="modelValue === opt ? 'bg-white/60 text-slate-800 dark:bg-white/10 dark:text-slate-100' : 'text-slate-600 dark:text-slate-300'"
-          >
-            <i v-if="icon" class="fa-solid text-xs opacity-70" :class="icon(opt)"></i>
-            <span>{{ opt }}</span>
-            <i v-if="modelValue === opt" class="fa-solid fa-check ml-auto text-[10px]"></i>
-          </button>
-        </div>
-      </div>
-    </Transition>
-  </div>
-</template>
-
-<style scoped>
-.dropdown-enter-active,
-.dropdown-leave-active {
-  transition: all 0.25s cubic-bezier(0.16, 1, 0.3, 1);
-}
-.dropdown-enter-from,
-.dropdown-leave-to {
-  opacity: 0;
-  transform: translateY(-8px) scale(0.98);
-}
-</style>
diff --git a/frontend/src/components/DateRangePicker.vue b/frontend/src/components/DateRangePicker.vue
deleted file mode 100644
index 535d0ba2..00000000
--- a/frontend/src/components/DateRangePicker.vue
+++ /dev/null
@@ -1,223 +0,0 @@
-<script setup>
-import { ref, computed } from 'vue'
-import { useClickOutside } from '../composables/useClickOutside.js'
-
-const props = defineProps({
-  startDate: { type: String, default: '' },
-  endDate: { type: String, default: '' },
-  label: { type: String, default: '时间跨度' },
-})
-
-const emit = defineEmits(['update:startDate', 'update:endDate'])
-
-const open = ref(false)
-const hoverDate = ref('')
-
-const calYear = ref(new Date().getFullYear())
-const calMonth = ref(new Date().getMonth())
-
-const WEEKDAYS = ['一', '二', '三', '四', '五', '六', '日']
-
-const toggle = () => {
-  if (open.value) { open.value = false; return }
-  const d = props.startDate ? new Date(props.startDate) : new Date()
-  calYear.value = d.getFullYear()
-  calMonth.value = d.getMonth()
-  open.value = true
-}
-
-const prevMonth = () => {
-  if (calMonth.value === 0) { calMonth.value = 11; calYear.value-- }
-  else calMonth.value--
-}
-
-const nextMonth = () => {
-  if (calMonth.value === 11) { calMonth.value = 0; calYear.value++ }
-  else calMonth.value++
-}
-
-const buildDays = (year, month) => {
-  let firstDay = new Date(year, month, 1).getDay()
-  let dow = firstDay === 0 ? 6 : firstDay - 1
-  const dim = new Date(year, month + 1, 0).getDate()
-  const prevDim = new Date(year, month, 0).getDate()
-  const days = []
-  
-  for (let i = dow - 1; i >= 0; i--) {
-    days.push({ day: prevDim - i, cur: false, month: month === 0 ? 11 : month - 1, year: month === 0 ? year - 1 : year })
-  }
-  for (let d = 1; d <= dim; d++) {
-    days.push({ day: d, cur: true, month, year })
-  }
-  while (days.length < 42) {
-    const d = days.length - dow - dim + 1
-    days.push({ day: d, cur: false, month: month === 11 ? 0 : month + 1, year: month === 11 ? year + 1 : year })
-  }
-  return days
-}
-
-const calDays = computed(() => buildDays(calYear.value, calMonth.value))
-
-const toStr = (d) =>
-  `${d.year}-${String(d.month + 1).padStart(2, '0')}-${String(d.day).padStart(2, '0')}`
-
-const effectiveEnd = computed(() =>
-  props.endDate || (props.startDate && hoverDate.value ? hoverDate.value : ''))
-
-const isStart = (d) => toStr(d) === props.startDate
-const isEnd = (d) => toStr(d) === props.endDate
-
-const rangeState = (d) => {
-  const s = toStr(d)
-  const lo = props.startDate
-  const hi = effectiveEnd.value
-  if (!lo) return null
-  
-  const [from, to] = lo <= hi ? [lo, hi] : [hi, lo]
-  
-  if (s === from && s === to) return 'single'
-  if (s === from) return 'start'
-  if (s === to) return 'end'
-  if (s > from && s < to) return 'mid'
-  return null
-}
-
-const isToday = (d) => {
-  const n = new Date()
-  return d.day === n.getDate() && d.month === n.getMonth() && d.year === n.getFullYear()
-}
-
-const triggerLabel = computed(() => {
-  if (props.startDate && props.endDate) {
-    if (props.startDate === props.endDate) return props.startDate
-    return `${props.startDate} — ${props.endDate}`
-  }
-  if (props.startDate) return `${props.startDate} — ...`
-  return ''
-})
-
-const selectDay = (d) => {
-  if (!d.cur) return
-  const s = toStr(d)
-  if (!props.startDate || (props.startDate && props.endDate)) {
-    emit('update:startDate', s)
-    emit('update:endDate', '')
-  } else {
-    if (s < props.startDate) {
-      emit('update:endDate', props.startDate)
-      emit('update:startDate', s)
-    } else {
-      emit('update:endDate', s)
-    }
-    hoverDate.value = ''
-    // 选择完范围后自动关闭
-    setTimeout(() => { open.value = false }, 150)
-  }
-}
-
-const clear = () => {
-  emit('update:startDate', '')
-  emit('update:endDate', '')
-}
-
-const selectToday = () => {
-  const n = new Date()
-  const s = `${n.getFullYear()}-${String(n.getMonth() + 1).padStart(2, '0')}-${String(n.getDate()).padStart(2, '0')}`
-  emit('update:startDate', s)
-  emit('update:endDate', s)
-  setTimeout(() => { open.value = false }, 150)
-}
-
-useClickOutside('.date-range-wrapper', () => { open.value = false })
-</script>
-
-<template>
-  <div class="date-range-wrapper relative w-full max-w-xs">
-    <label v-if="label" class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500 dark:text-slate-500">{{ label }}</label>
-
-    <!-- 触发器 -->
-    <div @click.stop="toggle"
-      class="group flex h-11 w-full cursor-pointer items-center rounded-xl border bg-white/60 px-4 text-sm font-bold backdrop-blur-xl transition-all hover:border-slate-300 hover:bg-white/80 hover:shadow-sm dark:border-white/10 dark:bg-white/[0.03] dark:hover:border-white/20 dark:hover:shadow-white/5"
-      :class="open ? 'border-white/30 bg-white/80 shadow-sm ring-2 ring-white/10 dark:border-white/20 dark:bg-white/[0.06] dark:ring-white/5' : 'border-slate-200/60'"
-    >
-      <i class="fa-regular fa-calendar mr-2 text-sm text-slate-400 transition-colors group-hover:text-slate-500 dark:text-slate-500 dark:group-hover:text-slate-300"></i>
-      <span v-if="triggerLabel" class="truncate text-slate-800 dark:text-slate-200">{{ triggerLabel }}</span>
-      <span v-else class="truncate text-slate-400 dark:text-slate-500">选择日期范围</span>
-      <i v-if="props.startDate" @click.stop="clear" class="fa-solid fa-circle-xmark ml-auto text-slate-300 hover:text-rose-500 dark:text-slate-600 dark:hover:text-rose-400 transition-colors"></i>
-    </div>
-
-    <!-- 下拉日历 -->
-    <Transition name="cal-dropdown">
-      <div v-if="open"
-        class="absolute left-0 top-full z-50 mt-2 w-72 overflow-hidden rounded-2xl border border-slate-200/60 bg-white/95 p-4 shadow-xl backdrop-blur-3xl dark:border-white/10 dark:bg-[#12121A]/90 dark:bg-gradient-to-b dark:from-white/[0.08] dark:to-transparent dark:backdrop-blur-3xl dark:shadow-[0_20px_50px_rgba(0,0,0,0.6)]"
-      >
-        <!-- 月份导航 -->
-        <div class="mb-4 flex items-center justify-between">
-          <button @click.stop="prevMonth"
-            class="flex h-8 w-8 items-center justify-center rounded-lg text-slate-400 transition-colors hover:bg-slate-100 hover:text-slate-600 dark:hover:bg-white/5 dark:hover:text-white">
-            <i class="fa-solid fa-chevron-left text-xs"></i>
-          </button>
-          <span class="text-sm font-black text-slate-700 dark:text-white">{{ calYear }}年 {{ calMonth + 1 }}月</span>
-          <button @click.stop="nextMonth"
-            class="flex h-8 w-8 items-center justify-center rounded-lg text-slate-400 transition-colors hover:bg-slate-100 hover:text-slate-600 dark:hover:bg-white/5 dark:hover:text-white">
-            <i class="fa-solid fa-chevron-right text-xs"></i>
-          </button>
-        </div>
-
-        <!-- 星期标题 -->
-        <div class="mb-2 grid grid-cols-7">
-          <span v-for="w in WEEKDAYS" :key="w" class="py-1 text-center text-[10px] font-black text-slate-400 dark:text-slate-500">{{ w }}</span>
-        </div>
-
-        <!-- 日期网格 -->
-        <div class="grid grid-cols-7 gap-y-1">
-          <div v-for="(d, i) in calDays" :key="i" class="relative flex h-8 items-center justify-center">
-            <!-- 范围背景层 (shadcn/ui style) -->
-            <div v-if="rangeState(d)"
-              class="absolute inset-y-0.5 z-0 transition-all duration-150"
-              :class="{
-                'left-1/2 right-0 bg-slate-100 dark:bg-white/10': rangeState(d) === 'start',
-                'left-0 right-1/2 bg-slate-100 dark:bg-white/10': rangeState(d) === 'end',
-                'left-0 right-0 bg-slate-50 dark:bg-white/5': rangeState(d) === 'mid',
-                'inset-x-0 mx-auto w-8 rounded-lg bg-slate-900 dark:bg-white': rangeState(d) === 'single'
-              }"
-            ></div>
-
-            <button
-              @click.stop="selectDay(d)"
-              @mouseenter="d.cur && props.startDate && !props.endDate ? hoverDate = toStr(d) : null"
-              @mouseleave="hoverDate = ''"
-              class="relative z-10 flex h-8 w-8 items-center justify-center rounded-lg text-xs font-bold transition-all"
-              :class="[
-                !d.cur ? 'text-slate-300 dark:text-slate-700 pointer-events-none' : 'text-slate-700 dark:text-slate-200 cursor-pointer',
-                rangeState(d) === 'start' || rangeState(d) === 'end' || rangeState(d) === 'single' ? '!bg-slate-900 !text-white dark:!bg-white dark:!text-slate-900' : '',
-                !rangeState(d) && d.cur ? 'hover:bg-slate-100 dark:hover:bg-white/10' : '',
-                isToday(d) && !rangeState(d) ? 'ring-1 ring-slate-400 dark:ring-white/40' : ''
-              ]"
-            >
-              {{ d.day }}
-            </button>
-          </div>
-        </div>
-
-        <!-- 底部操作 -->
-        <div class="mt-4 flex items-center justify-between border-t border-slate-100 pt-3 dark:border-white/5">
-          <button @click.stop="selectToday" class="text-xs font-black text-slate-600 hover:text-slate-800 dark:text-slate-400 dark:hover:text-slate-200 transition-colors">今天</button>
-          <button @click.stop="clear" class="text-xs font-bold text-slate-400 hover:text-rose-500 dark:text-slate-500 dark:hover:text-rose-400 transition-colors">清除选择</button>
-        </div>
-      </div>
-    </Transition>
-  </div>
-</template>
-
-<style scoped>
-.cal-dropdown-enter-active,
-.cal-dropdown-leave-active {
-  transition: all 0.2s cubic-bezier(0.16, 1, 0.3, 1);
-}
-.cal-dropdown-enter-from,
-.cal-dropdown-leave-to {
-  opacity: 0;
-  transform: translateY(-6px);
-}
-</style>
diff --git a/frontend/src/components/FileList.vue b/frontend/src/components/FileList.vue
deleted file mode 100644
index d7d1dc76..00000000
--- a/frontend/src/components/FileList.vue
+++ /dev/null
@@ -1,70 +0,0 @@
-<script setup>
-defineProps({
-  pendingFiles: { type: Array, default: () => [] },
-  fileProgress: { type: Object, default: () => ({}) },
-  waitingKeys: { type: Object, default: () => new Set() },
-  uploadBusy: { type: Boolean, default: false },
-  uploadReady: { type: Boolean, default: false },
-  splitting: { type: Boolean, default: false },
-  splitCompleted: { type: Boolean, default: false },
-})
-const emit = defineEmits(['remove-file'])
-</script>
-
-<template>
-  <Transition
-    enter-active-class="transition-all duration-500 ease-out"
-    enter-from-class="opacity-0 -translate-y-4 max-h-0"
-    enter-to-class="opacity-100 translate-y-0 max-h-40"
-    leave-active-class="transition-all duration-300 ease-in"
-    leave-from-class="opacity-100 translate-y-0 max-h-40"
-    leave-to-class="opacity-0 -translate-y-4 max-h-0"
-  >
-    <div v-if="pendingFiles.length" class="relative z-20 w-full overflow-hidden shrink-0">
-      <div class="flex flex-wrap items-center gap-3">
-        <div
-          v-for="item in pendingFiles"
-          :key="item.key"
-          class="file-chip relative flex items-center gap-3 overflow-hidden rounded-xl border border-slate-900/10 bg-slate-900/[0.04] p-2.5 shadow-sm backdrop-blur-sm dark:border-white/5 dark:bg-white/5"
-          :class="(uploadBusy || uploadReady) && 'file-chip--progress'"
-        >
-          <!-- 进度条背景 -->
-          <div 
-            class="absolute inset-0 -z-10 bg-blue-500/10 transition-[width] duration-500 ease-out dark:bg-indigo-500/20"
-            :style="{ width: `${fileProgress[item.key] ?? 0}%` }"
-          ></div>
-
-          <div class="flex h-8 w-8 shrink-0 items-center justify-center rounded-lg border border-slate-900/5 bg-white/50 shadow-sm dark:bg-white/10">
-            <i
-              class="fa-solid text-base"
-              :class="
-                item.file.name.toLowerCase().endsWith('.pdf')
-                  ? 'fa-file-pdf text-rose-500'
-                  : 'fa-file-image text-blue-500'
-              "
-            ></i>
-          </div>
-          <div class="flex flex-col min-w-0 pr-6 text-left">
-            <span class="truncate text-xs font-bold text-slate-800 dark:text-slate-200" :title="item.file.name">
-              {{ item.file.name }}
-            </span>
-            <div class="flex items-center gap-2">
-              <span class="text-[9px] font-black uppercase tracking-wider text-slate-400">
-                {{ waitingKeys.has(item.key) && uploadBusy ? '等待' : `${Math.round(fileProgress[item.key] ?? 0)}%` }}
-              </span>
-            </div>
-          </div>
-          
-          <button
-            type="button"
-            class="absolute right-1 top-1 flex h-5 w-5 items-center justify-center rounded-md text-slate-400 transition-colors hover:bg-rose-50 hover:text-rose-500 dark:hover:bg-rose-500/10"
-            :disabled="splitting || splitCompleted"
-            @click.stop="() => emit('remove-file', item.key)"
-          >
-            <i class="fa-solid fa-xmark text-[10px]"></i>
-          </button>
-        </div>
-      </div>
-    </div>
-  </Transition>
-</template>
diff --git a/frontend/src/components/FileUploader.vue b/frontend/src/components/FileUploader.vue
deleted file mode 100644
index 329b9ae0..00000000
--- a/frontend/src/components/FileUploader.vue
+++ /dev/null
@@ -1,99 +0,0 @@
-<script setup>
-import { ref } from 'vue'
-
-const props = defineProps({
-  pendingFiles: { type: Array, default: () => [] },
-  fileProgress: { type: Object, default: () => ({}) },
-  waitingKeys: { type: Object, default: () => new Set() },
-  uploadBusy: { type: Boolean, default: false },
-  uploadReady: { type: Boolean, default: false },
-  splitting: { type: Boolean, default: false },
-  splitCompleted: { type: Boolean, default: false },
-  expand: { type: Boolean, default: false },
-  disabled: { type: Boolean, default: false },
-})
-
-const emit = defineEmits(['upload', 'remove-file'])
-
-const fileInputEl = ref(null)
-const uploadHover = ref(false)
-
-const onFileInput = (e) => {
-  if (props.disabled) return
-  emit('upload', e.target.files)
-  e.target.value = ''
-}
-
-const onDrop = (e) => {
-  e.preventDefault()
-  uploadHover.value = false
-  if (props.disabled) return
-  emit('upload', e.dataTransfer.files)
-}
-
-const onClickZone = () => {
-  if (props.disabled) return
-  fileInputEl.value?.click()
-}
-</script>
-
-<template>
-  <div :class="{ 'flex flex-col flex-1': expand }">
-    <!-- Drop zone (拖拽上传区域) -->
-    <div
-      class="group relative flex flex-col items-center justify-center overflow-hidden rounded-[2rem] border backdrop-blur-md"
-      :class="[
-        uploadHover
-          ? 'border-blue-500/50 bg-slate-900/10 shadow-2xl shadow-blue-500/20 dark:border-indigo-500/50 dark:bg-white/10'
-          : 'border-slate-900/10 bg-slate-900/[0.05] shadow-[0_8px_30px_rgba(0,0,0,0.06)] hover:border-blue-200 hover:bg-slate-900/10 dark:border-white/5 dark:bg-white/5 dark:shadow-none dark:hover:border-indigo-500/20 dark:hover:bg-white/10',
-        expand ? 'flex-1 py-12 min-h-[280px]' : 'py-10',
-        disabled ? 'opacity-50 cursor-not-allowed' : 'cursor-pointer'
-      ]"
-      @dragenter.prevent="uploadHover = true"
-      @dragover.prevent="uploadHover = true"
-      @dragleave.prevent="uploadHover = false"
-      @drop="onDrop"
-      @click="onClickZone"
-      @keydown.enter.prevent="onClickZone"
-      @keydown.space.prevent="onClickZone"
-      role="button"
-      tabindex="0"
-    >
-      <!-- 动态装饰 -->
-      <div class="pointer-events-none absolute inset-0 -z-10">
-        <div class="absolute -left-10 -top-10 h-24 w-24 rounded-full bg-blue-500/5 blur-2xl transition-opacity group-hover:opacity-100"></div>
-        <div class="absolute -right-10 -bottom-10 h-24 w-24 rounded-full bg-indigo-500/5 blur-2xl transition-opacity group-hover:opacity-100"></div>
-      </div>
-
-      <!-- Center content (主上传引导) -->
-      <div class="flex flex-1 flex-col items-center justify-center">
-        <div class="relative mb-6 flex h-16 w-16 items-center justify-center rounded-2xl border border-slate-900/10 bg-slate-900/5 shadow-xl shadow-slate-200/50 backdrop-blur-md transition-transform duration-500 group-hover:scale-110 dark:border-white/10 dark:bg-white/5 dark:shadow-none">
-          <i class="fa-solid fa-cloud-arrow-up text-2xl text-blue-600 dark:text-indigo-400"></i>
-        </div>
-
-        <div class="max-w-xs">
-          <h4 v-if="disabled" class="text-sm font-black tracking-tight text-slate-500 dark:text-slate-400">
-            请先在 <span class="text-blue-600 dark:text-indigo-400">系统设置</span> 中配置 API 模型
-          </h4>
-          <h4 v-else class="text-sm font-black tracking-tight text-slate-900 dark:text-white">
-            点击 <span class="text-blue-600 dark:text-indigo-400">选择文件</span> 或拖拽至此
-          </h4>
-          <div class="mt-4 flex flex-wrap justify-center gap-1.5">
-            <span v-for="fmt in ['PDF', 'PNG', 'JPG']" :key="fmt" class="rounded-lg border border-slate-900/10 bg-slate-900/5 px-3 py-1 text-[9px] font-black tracking-wider text-slate-500 shadow-sm backdrop-blur-sm dark:border-white/5 dark:bg-white/5 dark:text-slate-400">
-              {{ fmt }}
-            </span>
-          </div>
-        </div>
-      </div>
-
-      <input
-        ref="fileInputEl"
-        type="file"
-        class="hidden"
-        accept=".pdf,.png,.jpg,.jpeg,.bmp,.tiff,.webp"
-        multiple
-        @change="onFileInput"
-      />
-    </div>
-  </div>
-</template>
\ No newline at end of file
diff --git a/frontend/src/components/GlassCard.vue b/frontend/src/components/GlassCard.vue
deleted file mode 100644
index cf5022e3..00000000
--- a/frontend/src/components/GlassCard.vue
+++ /dev/null
@@ -1,15 +0,0 @@
-<script setup>
-defineProps({
-  padding: { type: String, default: 'p-6' },
-  rounded: { type: String, default: 'rounded-2xl' },
-})
-</script>
-
-<template>
-  <div
-    :class="[rounded, padding]"
-    class="border border-slate-200/60 bg-white/70 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]"
-  >
-    <slot />
-  </div>
-</template>
diff --git a/frontend/src/components/LoginView.vue b/frontend/src/components/LoginView.vue
deleted file mode 100644
index bceb1d9a..00000000
--- a/frontend/src/components/LoginView.vue
+++ /dev/null
@@ -1,86 +0,0 @@
-<script setup>
-import { ref, reactive } from 'vue'
-import { useRouter } from 'vue-router'
-import { useAuth } from '../composables/useAuth.js'
-
-const router = useRouter()
-const { currentUser } = useAuth()
-
-const loading = ref(false)
-const showPwd = ref(false)
-const error = ref('')
-const form = reactive({ email: '', password: '' })
-
-async function handleLogin() {
-  error.value = ''
-  loading.value = true
-  try {
-    const res = await fetch('/api/auth/login', {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ email: form.email, password: form.password }),
-    })
-    const data = await res.json()
-    if (!res.ok) {
-      error.value = data.error || '登录失败'
-    } else {
-      currentUser.value = data.user
-      router.push('/app')
-    }
-  } catch {
-    error.value = '网络错误，请重试'
-  } finally {
-    loading.value = false
-  }
-}
-</script>
-
-<template>
-  <form @submit.prevent="handleLogin" class="space-y-4">
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">邮箱或用户名</label>
-      <input
-        v-model="form.email"
-        type="text"
-        required
-        autocomplete="username"
-        placeholder="邮箱 或 用户名"
-        class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-      />
-    </div>
-
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">密码</label>
-      <div class="relative">
-        <input
-          v-model="form.password"
-          :type="showPwd ? 'text' : 'password'"
-          required
-          autocomplete="current-password"
-          placeholder="请输入密码"
-          class="w-full h-10 px-4 pr-11 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-        />
-        <button
-          type="button"
-          @click="showPwd = !showPwd"
-          class="absolute right-3 top-1/2 -translate-y-1/2 text-slate-400 hover:text-slate-600 dark:hover:text-slate-200 transition-colors"
-        >
-          <i :class="showPwd ? 'fas fa-eye-slash' : 'fas fa-eye'"></i>
-        </button>
-      </div>
-    </div>
-
-    <p v-if="error" class="text-sm text-rose-500 dark:text-rose-400 flex items-center gap-2">
-      <i class="fas fa-circle-exclamation"></i>{{ error }}
-    </p>
-
-    <button
-      type="submit"
-      :disabled="loading"
-      class="w-full h-12 bg-blue-600 hover:bg-blue-700 disabled:opacity-50 disabled:cursor-not-allowed text-white font-bold rounded-xl transition-all shadow-md shadow-blue-500/30 hover:shadow-blue-500/50 flex items-center justify-center gap-2 text-sm"
-    >
-      <i v-if="loading" class="fas fa-spinner fa-spin"></i>
-      <span>{{ loading ? '登录中...' : '登录' }}</span>
-    </button>
-  </form>
-</template>
diff --git a/frontend/src/components/PageHeader.vue b/frontend/src/components/PageHeader.vue
deleted file mode 100644
index 89b71d2e..00000000
--- a/frontend/src/components/PageHeader.vue
+++ /dev/null
@@ -1,59 +0,0 @@
-<script setup>
-/**
- * PageHeader.vue
- * 统一的页面顶部组件，支持徽章、标题、副标题以及右侧操作区插槽。
- * 采用了现代化的磨砂感设计和优雅的入场动画。
- */
-defineProps({
-  badge: { type: String, default: '' },
-  badgeIcon: { type: String, default: '' },
-  badgeColor: { type: String, default: 'indigo' }, // 支持 indigo, blue, emerald, amber, rose, slate
-  title: { type: String, required: true },
-  subtitle: { type: String, default: '' },
-  subtitleIcon: { type: String, default: '' },
-})
-
-// 颜色映射，确保 Tailwind JIT 能识别并生成对应的样式
-const colorMap = {
-  indigo: 'border-indigo-200 bg-indigo-50 text-indigo-600 dark:border-indigo-500/30 dark:bg-indigo-500/10 dark:text-indigo-300 shadow-indigo-500/10',
-  blue: 'border-blue-200 bg-blue-50 text-blue-600 dark:border-blue-500/30 dark:bg-blue-500/10 dark:text-blue-300 shadow-blue-500/10',
-  emerald: 'border-emerald-200 bg-emerald-50 text-emerald-600 dark:border-emerald-500/30 dark:bg-emerald-500/10 dark:text-emerald-300 shadow-emerald-500/10',
-  amber: 'border-amber-200 bg-amber-50 text-amber-600 dark:border-amber-500/30 dark:bg-amber-500/10 dark:text-amber-300 shadow-amber-500/10',
-  rose: 'border-rose-200 bg-rose-50 text-rose-600 dark:border-rose-500/30 dark:bg-rose-500/10 dark:text-rose-300 shadow-rose-500/10',
-  slate: 'border-slate-200 bg-slate-50 text-slate-600 dark:border-slate-500/30 dark:bg-slate-500/10 dark:text-slate-300 shadow-slate-500/10',
-}
-</script>
-
-<template>
-  <header class="page-header relative z-30 mb-8 flex flex-col gap-6 sm:flex-row sm:items-end sm:justify-between">
-    <div class="flex flex-col items-start space-y-2">
-      <!-- 顶部徽章 -->
-      <div
-        v-if="badge"
-        class="inline-flex items-center gap-2 rounded-full border px-3 py-1 text-xs font-black uppercase tracking-widest shadow-sm"
-        :class="colorMap[badgeColor] || colorMap.indigo"
-      >
-        <i v-if="badgeIcon" :class="badgeIcon" class="animate-pulse"></i>
-        <span>{{ badge }}</span>
-      </div>
-
-      <div class="space-y-1">
-        <!-- 标题：大而有力的排版，增强视觉重心 -->
-        <h2 class="text-3xl font-black tracking-tight text-slate-900 sm:text-4xl dark:text-white">
-          {{ title }}
-        </h2>
-
-        <!-- 副标题：柔和的辅助信息 -->
-        <p v-if="subtitle" class="flex items-center gap-2 text-sm font-bold text-slate-500 dark:text-slate-400">
-          <i v-if="subtitleIcon" :class="subtitleIcon" class="opacity-60"></i>
-          <span>{{ subtitle }}</span>
-        </p>
-      </div>
-    </div>
-
-    <!-- 右侧操作区：通过 extra 插槽注入按钮、搜索框或筛选器 -->
-    <div v-if="$slots.extra" class="flex shrink-0 items-center gap-4">
-      <slot name="extra"></slot>
-    </div>
-  </header>
-</template>
diff --git a/frontend/src/components/RegisterView.vue b/frontend/src/components/RegisterView.vue
deleted file mode 100644
index 55288fb1..00000000
--- a/frontend/src/components/RegisterView.vue
+++ /dev/null
@@ -1,125 +0,0 @@
-<script setup>
-import { ref, reactive } from 'vue'
-import { useRouter } from 'vue-router'
-import { useAuth } from '../composables/useAuth.js'
-
-const router = useRouter()
-const { currentUser } = useAuth()
-
-const loading = ref(false)
-const showPwd = ref(false)
-const error = ref('')
-const success = ref('')
-const form = reactive({ username: '', email: '', password: '', confirm: '' })
-
-const passwordMismatch = () => form.confirm && form.password !== form.confirm
-
-async function handleRegister() {
-  error.value = ''
-  success.value = ''
-  if (passwordMismatch()) { error.value = '两次密码不一致'; return }
-  loading.value = true
-  try {
-    const res = await fetch('/api/auth/register', {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ username: form.username, email: form.email, password: form.password }),
-    })
-    const data = await res.json()
-    if (!res.ok) {
-      error.value = data.error || '注册失败'
-    } else {
-      success.value = '注册成功！正在登录...'
-      currentUser.value = data.user
-      router.push('/app')
-    }
-  } catch {
-    error.value = '网络错误，请重试'
-  } finally {
-    loading.value = false
-  }
-}
-</script>
-
-<template>
-  <form @submit.prevent="handleRegister" class="space-y-4">
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">用户名</label>
-      <input
-        v-model="form.username"
-        type="text"
-        required
-        autocomplete="username"
-        placeholder="您的昵称"
-        maxlength="50"
-        class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-      />
-    </div>
-
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">邮箱</label>
-      <input
-        v-model="form.email"
-        type="email"
-        required
-        autocomplete="email"
-        placeholder="your@email.com"
-        class="w-full h-10 px-4 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-      />
-    </div>
-
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">密码</label>
-      <div class="relative">
-        <input
-          v-model="form.password"
-          :type="showPwd ? 'text' : 'password'"
-          required
-          autocomplete="new-password"
-          placeholder="至少 6 位"
-          minlength="6"
-          class="w-full h-10 px-4 pr-11 rounded-xl border border-slate-200/60 dark:border-white/10 bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500 dark:focus:border-blue-500 transition-all text-sm"
-        />
-        <button
-          type="button"
-          @click="showPwd = !showPwd"
-          class="absolute right-3 top-1/2 -translate-y-1/2 text-slate-400 hover:text-slate-600 dark:hover:text-slate-200 transition-colors"
-        >
-          <i :class="showPwd ? 'fas fa-eye-slash' : 'fas fa-eye'"></i>
-        </button>
-      </div>
-    </div>
-
-    <div>
-      <label class="block text-sm font-medium text-slate-700 dark:text-slate-300 mb-2">确认密码</label>
-      <input
-        v-model="form.confirm"
-        type="password"
-        required
-        autocomplete="new-password"
-        placeholder="再次输入密码"
-        class="w-full h-10 px-4 rounded-xl border transition-all text-sm bg-white/80 dark:bg-white/5 text-slate-800 dark:text-white placeholder-slate-400 dark:placeholder-slate-500 focus:outline-none focus:ring-2 focus:ring-blue-500/50"
-        :class="passwordMismatch()
-          ? 'border-rose-400 dark:border-rose-500'
-          : 'border-slate-200/60 dark:border-white/10 focus:border-blue-500 dark:focus:border-blue-500'"
-      />
-      <p v-if="passwordMismatch()" class="text-xs text-rose-500 mt-1">两次密码不一致</p>
-    </div>
-
-    <p v-if="error" class="text-sm text-rose-500 dark:text-rose-400 flex items-center gap-2">
-      <i class="fas fa-circle-exclamation"></i>{{ error }}
-    </p>
-    <p v-if="success" class="text-sm text-emerald-600 dark:text-emerald-400 flex items-center gap-2">
-      <i class="fas fa-circle-check"></i>{{ success }}
-    </p>
-
-    <button
-      type="submit"
-      :disabled="loading || passwordMismatch()"
-      class="w-full h-12 bg-blue-600 hover:bg-blue-700 disabled:opacity-50 disabled:cursor-not-allowed text-white font-bold rounded-xl transition-all shadow-md shadow-blue-500/30 hover:shadow-blue-500/50 flex items-center justify-center gap-2 text-sm"
-    >
-      <i v-if="loading" class="fas fa-spinner fa-spin"></i>
-      <span>{{ loading ? '注册中...' : '创建账户' }}</span>
-    </button>
-  </form>
-</template>
diff --git a/frontend/src/components/SearchInput.vue b/frontend/src/components/SearchInput.vue
deleted file mode 100644
index 6903f127..00000000
--- a/frontend/src/components/SearchInput.vue
+++ /dev/null
@@ -1,26 +0,0 @@
-<script setup>
-defineProps({
-  modelValue: { type: String, default: '' },
-  label: { type: String, default: '' },
-  placeholder: { type: String, default: '搜索...' },
-  icon: { type: String, default: 'fa-solid fa-magnifying-glass' },
-})
-
-const emit = defineEmits(['update:modelValue'])
-</script>
-
-<template>
-  <div>
-    <label v-if="label" class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500 dark:text-slate-500">{{ label }}</label>
-    <div class="relative group">
-      <i :class="icon" class="pointer-events-none absolute left-4 top-1/2 z-10 -translate-y-1/2 text-sm text-slate-400 transition-colors group-hover:text-slate-500 group-focus-within:text-slate-600 dark:text-slate-500 dark:group-hover:text-slate-300 dark:group-focus-within:text-slate-200"></i>
-      <input
-        :value="modelValue"
-        @input="emit('update:modelValue', $event.target.value)"
-        type="text"
-        :placeholder="placeholder"
-        class="h-11 w-full rounded-xl border border-slate-200/60 bg-white/60 pl-11 pr-4 text-sm font-bold backdrop-blur-xl outline-none transition-all hover:border-slate-300 hover:bg-white/80 hover:shadow-sm focus:border-white/30 focus:bg-white/80 focus:shadow-sm focus:ring-2 focus:ring-white/10 dark:border-white/10 dark:bg-white/[0.03] dark:text-slate-200 dark:placeholder:text-slate-500 dark:hover:border-white/20 dark:hover:shadow-white/5 dark:focus:border-white/20 dark:focus:bg-white/[0.06] dark:focus:shadow-white/5 dark:focus:ring-white/5"
-      />
-    </div>
-  </div>
-</template>
diff --git a/frontend/src/components/StepIndicator.vue b/frontend/src/components/StepIndicator.vue
deleted file mode 100644
index c3030917..00000000
--- a/frontend/src/components/StepIndicator.vue
+++ /dev/null
@@ -1,84 +0,0 @@
-<script setup>
-defineProps({
-  step: { type: Number, default: 1 },
-})
-
-const labels = ['上传文件', 'OCR解析', '分割题目', '预览导出']
-const descs = ['选择 PDF 或图片', '结构化提取题干', 'AI 识别并拆分', '勾选题目并导出']
-</script>
-
-<template>
-  <div class="px-2 sm:px-0">
-    <ol class="relative flex flex-col gap-6 sm:flex-row sm:items-start sm:justify-between sm:gap-0">
-      <li 
-        v-for="n in 4" 
-        :key="n" 
-        class="group relative flex items-center gap-4 sm:flex-1 sm:flex-col sm:gap-2"
-      >
-        <!-- 桌面端连接线 (优化定位，避免穿过圆圈) -->
-        <div 
-          v-if="n < 4" 
-          class="hidden sm:block absolute left-[calc(50%+1.25rem)] top-5 w-[calc(100%-2.5rem)] h-[2px] -translate-y-1/2"
-          :class="n < step ? 'bg-slate-400 dark:bg-slate-600' : 'bg-slate-200 dark:bg-slate-800'"
-        ></div>
-        
-        <!-- 移动端连接线 -->
-        <div 
-          v-if="n < 4" 
-          class="sm:hidden absolute left-5 top-12 bottom-[-1.5rem] w-[2px] -translate-x-1/2"
-          :class="n < step ? 'bg-slate-400 dark:bg-slate-600' : 'bg-slate-200 dark:bg-slate-800'"
-        ></div>
-
-        <!-- 步骤圆圈 (白/灰黑玻璃风格) -->
-        <div 
-          class="step-circle relative z-10 flex h-10 w-10 shrink-0 items-center justify-center rounded-xl border backdrop-blur-sm"
-          :class="
-            n < step
-              ? 'border-transparent bg-gradient-to-br from-blue-600 to-indigo-600 text-white shadow-lg shadow-blue-500/30'
-              : n === step
-                ? 'border-slate-200 bg-white text-blue-600 shadow-xl shadow-blue-500/10 dark:border-white/20 dark:bg-white/10 dark:text-indigo-400'
-                : 'border-slate-200 bg-white text-slate-400 dark:border-white/5 dark:bg-white/5 dark:text-slate-600'
-          "
-        >
-          <Transition name="scale" mode="out-in">
-            <i v-if="n < step" :key="'check-'+n" class="fa-solid fa-check text-[10px] font-black"></i>
-            <span v-else :key="'num-'+n" class="text-xs font-black tabular-nums">{{ n }}</span>
-          </Transition>
-          
-          <!-- 当前步骤的光晕 -->
-          <div v-if="n === step" class="absolute -inset-1 animate-pulse rounded-xl bg-blue-500/10 blur-md dark:bg-indigo-500/20"></div>
-        </div>
-
-        <!-- 文本描述 (字号微调) -->
-        <div class="flex flex-col sm:items-center sm:text-center">
-          <h3 
-            class="text-[11px] font-black tracking-tight"
-            :class="n <= step ? 'text-slate-900 dark:text-white' : 'text-slate-500 dark:text-slate-600'"
-          >
-            {{ labels[n - 1] }}
-          </h3>
-          <p class="mt-0.5 text-[9px] font-bold leading-relaxed text-slate-500 dark:text-slate-500 sm:max-w-[100px]">
-            {{ descs[n - 1] }}
-          </p>
-        </div>
-      </li>
-    </ol>
-  </div>
-</template>
-
-<style scoped>
-.scale-enter-active,
-.scale-leave-active {
-  transition: all 0.3s cubic-bezier(0.34, 1.56, 0.64, 1);
-}
-
-.scale-enter-from {
-  opacity: 0;
-  transform: scale(0.5);
-}
-
-.scale-leave-to {
-  opacity: 0;
-  transform: scale(1.5);
-}
-</style>
\ No newline at end of file
diff --git a/frontend/src/components/auth/ForgotPasswordModal.vue b/frontend/src/components/auth/ForgotPasswordModal.vue
new file mode 100644
index 00000000..d08b4cf1
--- /dev/null
+++ b/frontend/src/components/auth/ForgotPasswordModal.vue
@@ -0,0 +1,231 @@
+<script setup>
+/**
+ * ForgotPasswordModal.vue
+ * 找回密码弹窗（邮箱验证码 + 重置密码）
+ */
+import { ref, reactive, onUnmounted, watch } from 'vue'
+import BaseButton from '@/components/base/BaseButton.vue'
+
+const props = defineProps({
+  open: { type: Boolean, default: false },
+})
+const emit = defineEmits(['close'])
+
+const step = ref(1)        // 1: 表单  2: 完成
+const loading = ref(false)
+const error = ref('')
+const showPwd = ref(false)
+const showConfirm = ref(false)
+const form = reactive({ email: '', code: '', password: '', confirm: '' })
+
+// 验证码倒计时
+const codeSending = ref(false)
+const countdown = ref(0)
+let timer = null
+onUnmounted(() => clearInterval(timer))
+
+// 打开时重置状态
+watch(() => props.open, (val) => {
+  if (val) {
+    step.value = 1
+    error.value = ''
+    showPwd.value = false
+    showConfirm.value = false
+    Object.assign(form, { email: '', code: '', password: '', confirm: '' })
+    countdown.value = 0
+    clearInterval(timer)
+  }
+})
+
+function startCountdown() {
+  countdown.value = 60
+  timer = setInterval(() => {
+    countdown.value--
+    if (countdown.value <= 0) clearInterval(timer)
+  }, 1000)
+}
+
+async function sendCode() {
+  if (codeSending.value || countdown.value > 0) return
+  const email = form.email.trim()
+  if (!email || !email.includes('@')) {
+    error.value = '请输入正确的邮箱'
+    return
+  }
+  error.value = ''
+  codeSending.value = true
+  try {
+    const res = await fetch('/api/auth/send-code', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ email, type: 'reset' }),
+    })
+    const data = await res.json()
+    if (!res.ok) {
+      error.value = data.error || '发送失败'
+    } else {
+      startCountdown()
+    }
+  } catch {
+    error.value = '网络错误，请重试'
+  } finally {
+    codeSending.value = false
+  }
+}
+
+async function resetPassword() {
+  error.value = ''
+  if (!form.email.trim() || !form.email.includes('@')) { error.value = '请输入正确的邮箱'; return }
+  if (!form.code.trim()) { error.value = '请输入验证码'; return }
+  if (form.password.length < 6) { error.value = '密码至少 6 位'; return }
+  if (form.password !== form.confirm) { error.value = '两次密码不一致'; return }
+  loading.value = true
+  try {
+    const res = await fetch('/api/auth/reset-password', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        email: form.email,
+        code: form.code.trim(),
+        password: form.password,
+      }),
+    })
+    const data = await res.json()
+    if (!res.ok) {
+      error.value = data.error || '重置失败'
+    } else {
+      step.value = 2
+    }
+  } catch {
+    error.value = '网络错误，请重试'
+  } finally {
+    loading.value = false
+  }
+}
+</script>
+
+<template>
+  <Teleport to="body">
+    <Transition name="fp-fade">
+      <div v-if="open" class="fixed inset-0 z-50 flex items-center justify-center p-4">
+        <!-- 遮罩 -->
+        <div class="absolute inset-0 bg-black/60 backdrop-blur-sm" @click="emit('close')"></div>
+
+        <!-- 弹窗主体 -->
+        <div
+          class="relative w-full max-w-sm rounded-2xl border border-white/[0.08] bg-[#0A0A0F]/95 backdrop-blur-xl p-6 shadow-xl"
+          @click.stop>
+          <!-- 关闭按钮 -->
+          <button @click="emit('close')"
+            class="absolute right-4 top-4 text-white/25 hover:text-white/50 transition-colors">
+            <i class="fas fa-xmark text-sm"></i>
+          </button>
+
+          <!-- Step 1: 邮箱 + 验证码 + 新密码 -->
+          <template v-if="step === 1">
+            <h3 class="text-xl font-bold text-white mb-1">找回密码</h3>
+            <p class="text-sm text-white/40 mb-6">输入注册邮箱和验证码，设置新密码</p>
+
+            <div class="space-y-4">
+              <div>
+                <label class="block text-sm font-medium text-white/60 mb-2">邮箱</label>
+                <div class="flex gap-2">
+                  <input v-model="form.email" type="email" placeholder="your@email.com"
+                    class="flex-1 min-w-0 h-10 px-4 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm" />
+                  <button type="button" @click="sendCode" :disabled="codeSending || countdown > 0"
+                    class="shrink-0 h-10 px-4 rounded-xl text-xs font-medium transition-all disabled:opacity-50 disabled:cursor-not-allowed"
+                    :class="countdown > 0
+                      ? 'bg-white/[0.03] text-white/30 border border-white/[0.06]'
+                      : 'bg-indigo-500/20 text-indigo-300 border border-indigo-500/20 hover:bg-indigo-500/30'">
+                    <i v-if="codeSending" class="fas fa-spinner fa-spin"></i>
+                    <template v-else-if="countdown > 0">{{ countdown }}s</template>
+                    <template v-else>发送验证码</template>
+                  </button>
+                </div>
+              </div>
+
+              <div>
+                <label class="block text-sm font-medium text-white/60 mb-2">验证码</label>
+                <input v-model="form.code" type="text" inputmode="numeric" maxlength="6" placeholder="请输入验证码"
+                  autocomplete="one-time-code"
+                  class="w-full h-10 px-4 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm tracking-widest" />
+              </div>
+
+              <div>
+                <label class="block text-sm font-medium text-white/60 mb-2">新密码</label>
+                <div class="relative">
+                  <input v-model="form.password" :type="showPwd ? 'text' : 'password'" placeholder="请输入新密码（至少 6 位）"
+                    class="w-full h-10 px-4 pr-11 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm" />
+                  <button type="button" @click="showPwd = !showPwd"
+                    class="absolute right-3 top-1/2 -translate-y-1/2 text-white/25 hover:text-white/50 transition-colors">
+                    <i :class="showPwd ? 'fas fa-eye-slash' : 'fas fa-eye'" class="text-xs"></i>
+                  </button>
+                </div>
+              </div>
+
+              <div>
+                <label class="block text-sm font-medium text-white/60 mb-2">确认密码</label>
+                <div class="relative">
+                  <input v-model="form.confirm" :type="showConfirm ? 'text' : 'password'" placeholder="再次输入新密码"
+                    class="w-full h-10 px-4 pr-11 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm" />
+                  <button type="button" @click="showConfirm = !showConfirm"
+                    class="absolute right-3 top-1/2 -translate-y-1/2 text-white/25 hover:text-white/50 transition-colors">
+                    <i :class="showConfirm ? 'fas fa-eye-slash' : 'fas fa-eye'" class="text-xs"></i>
+                  </button>
+                </div>
+              </div>
+
+              <p v-if="error" class="text-sm text-rose-400 flex items-center gap-2">
+                <i class="fas fa-circle-exclamation text-xs"></i>{{ error }}
+              </p>
+
+              <BaseButton variant="cta" class="w-full" :disabled="loading" @click="resetPassword">
+                <i v-if="loading" class="fas fa-spinner fa-spin text-xs"></i>
+                <span>{{ loading ? '重置中...' : '重置密码' }}</span>
+              </BaseButton>
+            </div>
+          </template>
+
+          <!-- Step 2: 完成 -->
+          <template v-else>
+            <div class="text-center py-4">
+              <div class="inline-flex items-center justify-center size-12 rounded-full bg-emerald-500/10 mb-4">
+                <i class="fas fa-check text-emerald-400 text-xl"></i>
+              </div>
+              <h3 class="text-xl font-bold text-white mb-2">密码已重置</h3>
+              <p class="text-sm text-white/40 mb-6">请使用新密码登录</p>
+              <BaseButton variant="cta" class="w-full" @click="emit('close')">
+                返回登录
+              </BaseButton>
+            </div>
+          </template>
+        </div>
+      </div>
+    </Transition>
+  </Teleport>
+</template>
+
+<style scoped>
+.fp-fade-enter-active,
+.fp-fade-leave-active {
+  transition: opacity 0.2s ease;
+}
+
+.fp-fade-enter-active .relative,
+.fp-fade-leave-active .relative {
+  transition: opacity 0.2s ease, transform 0.2s ease;
+}
+
+.fp-fade-enter-from,
+.fp-fade-leave-to {
+  opacity: 0;
+}
+
+.fp-fade-enter-from .relative {
+  transform: scale(0.95) translateY(8px);
+}
+
+.fp-fade-leave-to .relative {
+  transform: scale(0.97) translateY(4px);
+}
+</style>
diff --git a/frontend/src/components/base/BaseButton.vue b/frontend/src/components/base/BaseButton.vue
new file mode 100644
index 00000000..90bd5e2a
--- /dev/null
+++ b/frontend/src/components/base/BaseButton.vue
@@ -0,0 +1,104 @@
+<script setup>
+/**
+ * BaseButton.vue
+ * 品牌按钮（primary/secondary/cta 变体）
+ */
+import { computed } from 'vue'
+
+const props = defineProps({
+  variant: { type: String, default: 'primary' },   // primary | secondary | cta
+  size: { type: String, default: 'md' },            // sm | md
+  to: { type: String, default: '' },                // RouterLink 目标
+  href: { type: String, default: '' },              // 普通链接
+  type: { type: String, default: 'button' },        // button | submit
+  disabled: { type: Boolean, default: false },
+})
+
+const tag = computed(() => {
+  if (props.to) return 'RouterLink'
+  if (props.href) return 'a'
+  return 'button'
+})
+
+const bindProps = computed(() => {
+  if (props.to) return { to: props.to }
+  if (props.href) return { href: props.href }
+  return { type: props.type, disabled: props.disabled }
+})
+</script>
+
+<template>
+  <component
+    :is="tag"
+    v-bind="bindProps"
+    :class="[
+      'home-btn relative overflow-hidden inline-flex items-center justify-center font-medium gap-2 transition-all duration-200 disabled:opacity-50 disabled:cursor-not-allowed',
+      size === 'sm' ? 'h-8 px-4 text-xs rounded-lg' : 'h-10 px-6 text-sm rounded-xl',
+      {
+        'home-btn--primary': variant === 'primary',
+        'home-btn--secondary': variant === 'secondary',
+        'home-btn--cta': variant === 'cta',
+      },
+    ]"
+  >
+    <!-- 内部网格纹理 -->
+    <span v-if="variant !== 'secondary'" class="btn-grid" aria-hidden="true"></span>
+    <slot />
+  </component>
+</template>
+
+<style scoped>
+/* ── Primary: 品牌色渐变，微光 hover ── */
+.home-btn--primary {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  color: #fff;
+  border: none;
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+.home-btn--primary:hover {
+  background: linear-gradient(to bottom, rgba(145, 132, 235, 0.95), rgba(113, 100, 212, 0.95));
+  box-shadow:
+    inset 0 1px 0 0 rgba(255, 255, 255, 0.15),
+    0 0 20px 0 rgba(129, 115, 223, 0.25);
+}
+
+/* ── Secondary: 白玻璃按钮（与 brand-btn 一致） ── */
+.home-btn--secondary {
+  background: rgba(255, 255, 255, 0.05);
+  color: rgba(255, 255, 255, 0.55);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-top-color: rgba(255, 255, 255, 0.15);
+  border-bottom-color: rgba(255, 255, 255, 0.03);
+}
+.home-btn--secondary:hover {
+  background: rgba(255, 255, 255, 0.08);
+  color: rgba(255, 255, 255, 0.8);
+}
+
+/* ── CTA: 更强的品牌色，用于转化区 ── */
+.home-btn--cta {
+  background: linear-gradient(to bottom, rgba(109, 92, 214, 1), rgba(88, 72, 194, 1));
+  color: #fff;
+  border: none;
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+.home-btn--cta:hover {
+  background: linear-gradient(to bottom, rgba(125, 108, 228, 1), rgba(102, 86, 208, 1));
+  box-shadow:
+    inset 0 1px 0 0 rgba(255, 255, 255, 0.15),
+    0 0 24px 0 rgba(109, 92, 214, 0.3);
+}
+
+/* ── 按钮内网格纹理 ── */
+.btn-grid {
+  position: absolute;
+  inset: 0;
+  pointer-events: none;
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+</style>
diff --git a/frontend/src/components/base/BaseCard.vue b/frontend/src/components/base/BaseCard.vue
new file mode 100644
index 00000000..e17df68e
--- /dev/null
+++ b/frontend/src/components/base/BaseCard.vue
@@ -0,0 +1,19 @@
+<script setup>
+/**
+ * BaseCard.vue
+ * 毛玻璃风格卡片容器
+ */
+defineProps({
+  padding: { type: String, default: 'p-4' },
+  rounded: { type: String, default: 'rounded-lg' },
+})
+</script>
+
+<template>
+  <div
+    :class="[rounded, padding]"
+    class="border border-white/[0.06] bg-white/[0.02] transition-all hover:bg-white/[0.04] hover:border-white/[0.1]"
+  >
+    <slot />
+  </div>
+</template>
diff --git a/frontend/src/components/GlassButton.vue b/frontend/src/components/base/BaseGhostButton.vue
similarity index 60%
rename from frontend/src/components/GlassButton.vue
rename to frontend/src/components/base/BaseGhostButton.vue
index 935333ec..2c2a5e9a 100644
--- a/frontend/src/components/GlassButton.vue
+++ b/frontend/src/components/base/BaseGhostButton.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * BaseGhostButton.vue
+ * 毛玻璃风格按钮
+ */
 defineProps({
   icon: { type: String, default: '' },
 })
@@ -7,7 +11,7 @@ defineProps({
 <template>
   <button
     v-bind="$attrs"
-    class="group inline-flex h-10 items-center justify-center gap-2 rounded-xl border border-slate-200/60 bg-white/60 px-4 text-sm font-bold text-slate-700 backdrop-blur-xl transition-all hover:border-slate-300 hover:bg-white/80 dark:border-white/10 dark:bg-white/[0.03] dark:text-slate-300 dark:hover:border-white/20 dark:hover:bg-white/[0.06]"
+    class="group inline-flex h-10 items-center justify-center gap-2 rounded-xl border border-slate-200/60 bg-white/60 px-4 text-sm font-bold text-slate-700 transition-all hover:border-slate-300 hover:bg-white/80 dark:border-white/10 dark:bg-white/[0.03] dark:text-slate-300 dark:hover:border-white/20 dark:hover:bg-white/[0.06]"
   >
     <i v-if="icon" class="fa-solid transition-transform group-hover:scale-110" :class="icon"></i>
     <slot />
diff --git a/frontend/src/components/AppLoading.vue b/frontend/src/components/base/BaseLoading.vue
similarity index 70%
rename from frontend/src/components/AppLoading.vue
rename to frontend/src/components/base/BaseLoading.vue
index 721056e1..9cce786c 100644
--- a/frontend/src/components/AppLoading.vue
+++ b/frontend/src/components/base/BaseLoading.vue
@@ -1,4 +1,10 @@
 <script setup>
+/**
+ * BaseLoading.vue
+ * 全局加载遮罩（BaseLogo + 进度条动画）
+ */
+import BaseLogo from '@/components/base/BaseLogo.vue'
+
 defineProps({
   visible: { type: Boolean, default: true },
 })
@@ -8,15 +14,10 @@ const emit = defineEmits(['after-enter'])
 <template>
   <Transition name="loading-fade" @after-enter="emit('after-enter')">
     <div v-if="visible" class="fixed inset-0 z-[200] flex flex-col items-center justify-center gap-8 bg-[#0A0A0F]">
-      <div class="relative">
-        <div class="absolute inset-0 rounded-2xl bg-indigo-500/40 blur-xl"></div>
-        <div class="relative bg-gradient-to-br from-indigo-500 to-indigo-700 p-4 rounded-2xl shadow-sm border border-white/10">
-          <img src="/logo.svg" class="w-8 h-8 brightness-0 invert" alt="logo" />
-        </div>
-      </div>
+      <BaseLogo size="lg" breathe />
       <div class="w-48">
         <div class="h-0.5 w-full rounded-full bg-white/10 overflow-hidden">
-          <div class="h-full rounded-full bg-gradient-to-r from-indigo-500 to-blue-400 loading-bar"></div>
+          <div class="h-full rounded-full bg-gradient-to-r from-[rgba(129,115,223,0.8)] to-[rgba(99,87,199,0.8)] loading-bar"></div>
         </div>
       </div>
     </div>
diff --git a/frontend/src/components/base/BaseLogo.vue b/frontend/src/components/base/BaseLogo.vue
new file mode 100644
index 00000000..b9a84264
--- /dev/null
+++ b/frontend/src/components/base/BaseLogo.vue
@@ -0,0 +1,72 @@
+<script setup>
+/**
+ * BaseLogo.vue
+ * 品牌 Logo（支持呼吸动画、多尺寸）
+ */
+defineProps({
+  size: { type: String, default: 'md' }, // sm | md | lg
+  breathe: { type: Boolean, default: false },
+})
+</script>
+
+<template>
+  <div
+    class="brand-logo relative overflow-hidden flex items-center justify-center"
+    :class="[
+      breathe ? 'brand-logo--breathe' : '',
+      {
+        'p-1.5 rounded-lg': size === 'sm',
+        'p-2 rounded-xl': size === 'md',
+        'p-3 rounded-2xl': size === 'lg',
+      },
+    ]"
+  >
+    <span class="brand-logo__grid absolute inset-0 pointer-events-none"></span>
+    <img
+      src="/logo.svg"
+      class="relative brightness-0 invert"
+      :class="{
+        'w-4 h-4': size === 'sm',
+        'w-6 h-6': size === 'md',
+        'w-8 h-8': size === 'lg',
+      }"
+      alt="logo"
+    />
+  </div>
+</template>
+
+<style scoped>
+.brand-logo {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+
+.brand-logo__grid {
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+
+/* 呼吸发光 */
+@keyframes brand-breathe {
+  0%, 100% {
+    box-shadow:
+      inset 0 1px 0 0 rgba(255, 255, 255, 0.12),
+      0 0 12px rgba(129, 115, 223, 0.3),
+      0 0 24px rgba(129, 115, 223, 0.15);
+  }
+  50% {
+    box-shadow:
+      inset 0 1px 0 0 rgba(255, 255, 255, 0.15),
+      0 0 24px rgba(129, 115, 223, 0.5),
+      0 0 48px rgba(129, 115, 223, 0.25);
+  }
+}
+
+.brand-logo--breathe {
+  animation: brand-breathe 3s ease-in-out infinite;
+}
+</style>
diff --git a/frontend/src/components/base/BaseSelect.vue b/frontend/src/components/base/BaseSelect.vue
new file mode 100644
index 00000000..8362fc1d
--- /dev/null
+++ b/frontend/src/components/base/BaseSelect.vue
@@ -0,0 +1,80 @@
+<script setup>
+/**
+ * BaseSelect.vue
+ * 自定义下拉选择器
+ */
+import { ref } from 'vue'
+import { useClickOutside } from '@/composables/useClickOutside.js'
+
+const props = defineProps({
+  modelValue: { type: String, default: '' },
+  options: { type: Array, default: () => [] },
+  label: { type: String, default: '' },
+  placeholder: { type: String, default: '全部' },
+  icon: { type: Function, default: null },
+  widthClass: { type: String, default: '' },
+})
+
+const emit = defineEmits(['update:modelValue'])
+const open = ref(false)
+
+useClickOutside('.custom-select-wrapper', () => { open.value = false })
+
+const toggle = () => { open.value = !open.value }
+const select = (val) => { emit('update:modelValue', val); open.value = false }
+</script>
+
+<template>
+  <div class="custom-select-wrapper relative" :class="widthClass">
+    <label v-if="label" class="mb-1.5 block text-xs font-medium text-[#62666d]">{{ label }}</label>
+    <button
+      type="button"
+      @click.stop="toggle"
+      class="flex h-9 w-full items-center justify-between rounded-md border border-white/[0.08] bg-white/[0.02] px-3 text-left text-sm font-medium transition-colors hover:border-white/[0.12]"
+      :class="[
+        open ? 'border-white/[0.15]' : '',
+        modelValue ? 'text-[#d0d6e0]' : 'text-[#62666d]',
+      ]"
+    >
+      <span class="truncate">{{ modelValue || placeholder }}</span>
+      <i class="fa-solid fa-chevron-down ml-2 text-[10px] text-[#62666d] transition-transform duration-200" :class="open ? 'rotate-180' : ''"></i>
+    </button>
+    <Transition name="dropdown">
+      <div v-if="open" class="absolute left-0 top-full z-50 mt-1 min-w-full w-max rounded-lg bg-[#1c1c20] py-1 shadow-lg shadow-black/40">
+        <div class="no-scrollbar max-h-56 overflow-y-auto">
+          <button
+            type="button"
+            @click.stop="select('')"
+            class="dropdown-item"
+            :class="modelValue === '' ? 'text-[#f7f8f8]' : ''"
+          >
+            <span>{{ placeholder }}</span>
+            <i v-if="modelValue === ''" class="fa-solid fa-check ml-auto text-[10px] text-[rgb(129,115,223)]"></i>
+          </button>
+          <button
+            v-for="opt in options" :key="opt"
+            type="button"
+            @click.stop="select(opt)"
+            class="dropdown-item"
+            :class="modelValue === opt ? 'text-[#f7f8f8]' : ''"
+          >
+            <span>{{ opt }}</span>
+            <i v-if="modelValue === opt" class="fa-solid fa-check ml-auto text-[10px] text-[rgb(129,115,223)]"></i>
+          </button>
+        </div>
+      </div>
+    </Transition>
+  </div>
+</template>
+
+<style scoped>
+.dropdown-enter-active,
+.dropdown-leave-active {
+  transition: all 0.15s ease-out;
+}
+.dropdown-enter-from,
+.dropdown-leave-to {
+  opacity: 0;
+  transform: translateY(-4px);
+}
+</style>
diff --git a/frontend/src/components/base/EmptyState.vue b/frontend/src/components/base/EmptyState.vue
new file mode 100644
index 00000000..c1019282
--- /dev/null
+++ b/frontend/src/components/base/EmptyState.vue
@@ -0,0 +1,31 @@
+<script setup>
+/**
+ * EmptyState.vue
+ * Linear 风格空状态组件 — 居中图标 + 标题 + 描述 + 操作按钮
+ */
+defineProps({
+  icon: { type: String, default: '' },
+  title: { type: String, required: true },
+  description: { type: String, default: '' },
+})
+</script>
+
+<template>
+  <div class="flex flex-1 flex-col items-center justify-center">
+    <!-- 图标 -->
+    <div v-if="icon" class="mb-6 flex h-16 w-16 items-center justify-center rounded-xl border border-white/[0.06] bg-white/[0.02]">
+      <i :class="icon" class="text-2xl text-[#62666d]"></i>
+    </div>
+
+    <!-- 标题 -->
+    <h3 class="text-base font-medium text-[#f7f8f8]">{{ title }}</h3>
+
+    <!-- 描述 -->
+    <p v-if="description" class="mt-2 max-w-sm text-center text-sm text-[#62666d] leading-relaxed">{{ description }}</p>
+
+    <!-- 操作按钮区 -->
+    <div v-if="$slots.default" class="mt-6 flex items-center gap-3">
+      <slot></slot>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/ImageModal.vue b/frontend/src/components/base/ImageModal.vue
similarity index 88%
rename from frontend/src/components/ImageModal.vue
rename to frontend/src/components/base/ImageModal.vue
index 74881f4d..edfb44d9 100644
--- a/frontend/src/components/ImageModal.vue
+++ b/frontend/src/components/base/ImageModal.vue
@@ -1,5 +1,9 @@
 <script setup>
-import { clampScale } from '../utils.js'
+/**
+ * ImageModal.vue
+ * 图片预览弹窗（缩放 + 拖拽）
+ */
+import { clampScale } from '@/utils.js'
 
 const props = defineProps({
   open: { type: Boolean, default: false },
diff --git a/frontend/src/components/base/SearchInput.vue b/frontend/src/components/base/SearchInput.vue
new file mode 100644
index 00000000..249619fb
--- /dev/null
+++ b/frontend/src/components/base/SearchInput.vue
@@ -0,0 +1,30 @@
+<script setup>
+/**
+ * SearchInput.vue
+ * 搜索输入框（带图标 + 清除按钮）
+ */
+defineProps({
+  modelValue: { type: String, default: '' },
+  label: { type: String, default: '' },
+  placeholder: { type: String, default: '搜索...' },
+  icon: { type: String, default: 'fa-solid fa-magnifying-glass' },
+})
+
+const emit = defineEmits(['update:modelValue'])
+</script>
+
+<template>
+  <div>
+    <label v-if="label" class="mb-1.5 block text-xs font-medium text-[#62666d]">{{ label }}</label>
+    <div class="relative group">
+      <i :class="icon" class="pointer-events-none absolute left-3 top-1/2 z-10 -translate-y-1/2 text-xs text-[#62666d] transition-colors group-focus-within:text-[#8a8f98]"></i>
+      <input
+        :value="modelValue"
+        @input="emit('update:modelValue', $event.target.value)"
+        type="text"
+        :placeholder="placeholder"
+        class="h-9 w-full rounded-md border border-white/[0.08] bg-white/[0.02] pl-9 pr-3 text-sm font-medium text-[#f7f8f8] placeholder-[#62666d] outline-none transition-colors hover:border-white/[0.12] focus:border-white/[0.15] border-top-color-[rgba(255,255,255,0.12)]"
+      />
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/ToastContainer.vue b/frontend/src/components/base/ToastContainer.vue
similarity index 95%
rename from frontend/src/components/ToastContainer.vue
rename to frontend/src/components/base/ToastContainer.vue
index af6c12a9..95f2f733 100644
--- a/frontend/src/components/ToastContainer.vue
+++ b/frontend/src/components/base/ToastContainer.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * ToastContainer.vue
+ * Toast 通知容器（成功/错误/信息）
+ */
 defineProps({
   toasts: { type: Array, default: () => [] },
 })
@@ -32,7 +36,7 @@ const onLeave = (el) => {
       <div
         v-for="t in toasts"
         :key="t.id"
-        class="pointer-events-auto relative flex min-w-[320px] max-w-md items-center gap-4 overflow-hidden rounded-[1.25rem] border px-5 py-4 shadow-[0_20px_50px_rgba(0,0,0,0.1)] backdrop-blur-md dark:shadow-[0_20px_50px_rgba(0,0,0,0.3)]"
+        class="pointer-events-auto relative flex min-w-[320px] max-w-md items-center gap-4 overflow-hidden rounded-[1.25rem] border px-5 py-4 shadow-[0_20px_50px_rgba(0,0,0,0.1)] dark:shadow-[0_20px_50px_rgba(0,0,0,0.3)]"
         :class="
           t.type === 'success'
             ? 'border-emerald-500/20 bg-emerald-50/90 text-emerald-900 dark:border-emerald-500/20 dark:bg-emerald-500/10 dark:text-emerald-300'
diff --git a/frontend/src/components/StatCard.vue b/frontend/src/components/dashboard/StatCard.vue
similarity index 94%
rename from frontend/src/components/StatCard.vue
rename to frontend/src/components/dashboard/StatCard.vue
index 73c5f507..dc262220 100644
--- a/frontend/src/components/StatCard.vue
+++ b/frontend/src/components/dashboard/StatCard.vue
@@ -4,7 +4,7 @@
  * 统计数字卡片组件，带数字滚动动画。
  */
 import { ref, watch, onMounted } from 'vue'
-import GlassCard from './GlassCard.vue'
+import BaseCard from '@/components/base/BaseCard.vue'
 
 const props = defineProps({
   label: { type: String, required: true },
@@ -55,7 +55,7 @@ watch(() => props.value, (v) => { animateTo(v) })
 </script>
 
 <template>
-  <GlassCard padding="p-6" class="stat-card group transition-all hover:shadow-md">
+  <BaseCard padding="p-6" class="stat-card group transition-all hover:shadow-md">
     <div class="flex items-center gap-4">
       <div
         class="flex size-12 shrink-0 items-center justify-center rounded-lg transition-transform duration-500 group-hover:scale-110"
@@ -78,6 +78,6 @@ watch(() => props.value, (v) => { animateTo(v) })
         </div>
       </div>
     </div>
-  </GlassCard>
+  </BaseCard>
 </template>
 
diff --git a/frontend/src/components/home/HomeFooter.vue b/frontend/src/components/home/HomeFooter.vue
new file mode 100644
index 00000000..d7de20e7
--- /dev/null
+++ b/frontend/src/components/home/HomeFooter.vue
@@ -0,0 +1,31 @@
+<script setup>
+/**
+ * HomeFooter.vue
+ * 落地页底部行动号召区
+ */
+import BaseButton from '@/components/base/BaseButton.vue'
+</script>
+
+<template>
+  <section id="cta" class="relative z-10 flex flex-col items-center justify-center py-32 px-6 overflow-hidden min-h-[70vh]">
+    <!-- 底部中心光晕 -->
+    <div class="absolute bottom-0 left-1/2 -translate-x-1/2 w-[80vw] h-[50vh] pointer-events-none z-0">
+      <div class="absolute inset-0 bg-indigo-500/10 blur-[120px] rounded-[100%] transform translate-y-1/2"></div>
+      <div class="absolute inset-0 bg-violet-500/5 blur-[150px] rounded-[100%] transform translate-y-1/2 scale-150"></div>
+    </div>
+
+    <!-- CTA -->
+    <div class="flex-1 flex flex-col items-center justify-center text-center relative z-10 w-full">
+      <h2 class="reveal text-4xl sm:text-5xl font-semibold tracking-tight mb-6 text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">即刻开启高效学习模式</h2>
+      <p class="reveal text-white/40 text-base mb-10 max-w-md mx-auto leading-relaxed">告别低效抄录，让 AI 成为你的超级助教。注册即享完整试卷解析体验。</p>
+      <div class="reveal">
+        <BaseButton variant="cta" to="/auth" class="shadow-[0_0_30px_rgba(129,115,223,0.3)] hover:shadow-[0_0_50px_rgba(129,115,223,0.5)] transform hover:-translate-y-1 transition-all duration-300">
+          免费创建错题库
+        </BaseButton>
+      </div>
+    </div>
+  </section>
+</template>
diff --git a/frontend/src/components/home/HomeHeader.vue b/frontend/src/components/home/HomeHeader.vue
new file mode 100644
index 00000000..8de6c411
--- /dev/null
+++ b/frontend/src/components/home/HomeHeader.vue
@@ -0,0 +1,58 @@
+<script setup>
+/**
+ * HomeHeader.vue
+ * 落地页顶部导航栏
+ */
+import { useTheme } from '@/composables/useTheme.js'
+import BaseButton from '@/components/base/BaseButton.vue'
+import BaseLogo from '@/components/base/BaseLogo.vue'
+
+const { isDark, toggleTheme } = useTheme()
+
+const props = defineProps({
+  navScrolled: Boolean,
+  activeSection: String,
+  sections: Array,
+})
+
+const emit = defineEmits(['scrollToSection'])
+</script>
+
+<template>
+  <nav
+    id="top-nav"
+    class="fixed top-0 left-0 w-full z-50 border-b transition-[border-color] duration-200"
+    :class="navScrolled ? 'border-white/[0.06] bg-[#0A0A0F]/80' : 'border-transparent'"
+  >
+    <div class="max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
+      <div class="flex justify-between items-center h-14">
+        <!-- Logo -->
+        <div class="flex items-center gap-2.5">
+          <BaseLogo size="sm" />
+          <span class="text-sm font-semibold text-white/90 tracking-wide">智卷错题本</span>
+        </div>
+
+        <!-- 中间导航 -->
+        <div class="hidden md:flex items-center gap-1">
+          <button
+            v-for="s in sections"
+            :key="s.id"
+            @click="emit('scrollToSection', s.id)"
+            class="px-3 py-1.5 text-[13px] font-medium rounded-md transition-colors"
+            :class="activeSection === s.id ? 'text-white bg-white/[0.06]' : 'text-white/40 hover:text-white/70'"
+          >{{ s.label }}</button>
+        </div>
+
+        <!-- 右侧操作 -->
+        <div class="flex items-center gap-3">
+          <a href="https://github.com/xiaozhejiya/error_correction" target="_blank" rel="noopener noreferrer" class="text-white/30 hover:text-white/60 transition-colors">
+            <i class="fa-brands fa-github text-base"></i>
+          </a>
+          <BaseButton to="/auth" size="sm">
+            进入工作台
+          </BaseButton>
+        </div>
+      </div>
+    </div>
+  </nav>
+</template>
diff --git a/frontend/src/components/home/HomePill.vue b/frontend/src/components/home/HomePill.vue
new file mode 100644
index 00000000..d0547aa5
--- /dev/null
+++ b/frontend/src/components/home/HomePill.vue
@@ -0,0 +1,29 @@
+<script setup>
+/**
+ * HomePill.vue
+ * 落地页标签徽章
+ */
+</script>
+
+<template>
+  <div class="relative overflow-hidden inline-flex items-center rounded-full px-4 py-1 text-xs font-medium text-white/90 pill-fill">
+    <span class="pill-grid absolute inset-0 pointer-events-none"></span>
+    <slot>AI 驱动 · 专为学生设计</slot>
+  </div>
+</template>
+
+<style scoped>
+.pill-fill {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+
+.pill-grid {
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+</style>
diff --git a/frontend/src/components/home/HomeSideNav.vue b/frontend/src/components/home/HomeSideNav.vue
new file mode 100644
index 00000000..394dc53b
--- /dev/null
+++ b/frontend/src/components/home/HomeSideNav.vue
@@ -0,0 +1,49 @@
+<script setup>
+/**
+ * HomeSideNav.vue
+ * 落地页右侧浮动导航（Section 圆点 + 回到顶部按钮）
+ */
+import { ArrowUp } from 'lucide-vue-next'
+
+const props = defineProps({
+  activeSection: String,
+  sections: Array,
+  showBackToTop: Boolean,
+})
+
+const emit = defineEmits(['scrollToSection', 'back-to-top'])
+</script>
+
+<template>
+  <!-- 右侧 Section 导航 -->
+  <nav id="section-nav" class="fixed right-4 md:right-6 top-1/2 -translate-y-1/2 z-40 flex flex-col items-center gap-3 py-4 px-1.5 rounded-lg brand-btn scale-90 md:scale-100">
+    <button
+      v-for="s in sections"
+      :key="s.id"
+      class="relative flex items-center justify-center w-6 h-6 focus:outline-none group"
+      :title="s.label"
+      @click="emit('scrollToSection', s.id)"
+    >
+      <div
+        class="w-1 h-1 rounded-full transition-all duration-200"
+        :class="activeSection === s.id
+          ? 'bg-white scale-150'
+          : 'bg-white/20 group-hover:bg-white/50'"
+      ></div>
+      <span class="absolute right-10 px-2 py-1 rounded-md bg-[#111118] border border-white/[0.06] text-[11px] font-medium text-white/70 whitespace-nowrap pointer-events-none opacity-0 translate-x-1 group-hover:opacity-100 group-hover:translate-x-0 transition-all duration-200">
+        {{ s.label }}
+      </span>
+    </button>
+  </nav>
+
+  <!-- 回到顶部按钮 -->
+  <button
+    id="back-to-top"
+    class="fixed bottom-8 right-8 z-50 flex h-10 w-10 items-center justify-center rounded-full brand-btn text-[#8a8f98] transition-[opacity,transform] duration-300 hover:text-[#f7f8f8]"
+    :style="{ opacity: showBackToTop ? '1' : '0', transform: showBackToTop ? 'translateY(0)' : 'translateY(12px)', pointerEvents: showBackToTop ? 'auto' : 'none' }"
+    aria-label="回到顶部"
+    @click="emit('back-to-top')"
+  >
+    <ArrowUp class="h-4 w-4" />
+  </button>
+</template>
diff --git a/frontend/src/components/landing/LandingBackToTop.vue b/frontend/src/components/landing/LandingBackToTop.vue
deleted file mode 100644
index 99e110da..00000000
--- a/frontend/src/components/landing/LandingBackToTop.vue
+++ /dev/null
@@ -1,22 +0,0 @@
-<script setup>
-import { ArrowUp } from 'lucide-vue-next'
-
-const props = defineProps({
-  visible: Boolean,
-})
-
-const emit = defineEmits(['click'])
-</script>
-
-<template>
-  <!-- 回到顶部按钮 -->
-  <button
-    id="back-to-top"
-    class="fixed bottom-8 right-8 z-50 flex h-10 w-10 items-center justify-center rounded-full border border-slate-200/60 bg-white/80 text-slate-500 shadow-md backdrop-blur-md transition-all duration-300 hover:bg-white hover:text-blue-600 hover:border-blue-300/60 hover:shadow-[0_0_16px_4px_rgba(59,130,246,0.35)] dark:border-white/10 dark:bg-white/5 dark:text-slate-400 dark:hover:bg-white/10 dark:hover:text-indigo-400 dark:hover:border-indigo-500/40 dark:hover:shadow-[0_0_16px_4px_rgba(99,102,241,0.5)]"
-    :style="{ opacity: visible ? '1' : '0', transform: visible ? 'translateY(0)' : 'translateY(12px)', pointerEvents: visible ? 'auto' : 'none' }"
-    aria-label="回到顶部"
-    @click="emit('click')"
-  >
-    <ArrowUp class="h-4 w-4" />
-  </button>
-</template>
diff --git a/frontend/src/components/landing/LandingCta.vue b/frontend/src/components/landing/LandingCta.vue
deleted file mode 100644
index 853e69b5..00000000
--- a/frontend/src/components/landing/LandingCta.vue
+++ /dev/null
@@ -1,44 +0,0 @@
-<script setup>
-</script>
-
-<template>
-  <!-- CTA + Footer section -->
-  <section id="cta" class="relative z-10 min-h-screen flex flex-col justify-between py-16 px-8 overflow-hidden bg-gradient-to-br from-blue-600 via-indigo-600 to-indigo-700 dark:from-indigo-950 dark:via-[#0A0A0F] dark:to-slate-950">
-    <!-- 装饰光晕 -->
-    <div class="pointer-events-none absolute inset-0 z-0">
-      <div class="absolute top-[-15%] left-[-8%] h-[500px] w-[500px] rounded-full bg-blue-400/20 dark:bg-indigo-500/10 blur-[130px]"></div>
-      <div class="absolute bottom-[-10%] right-[-5%] h-[400px] w-[400px] rounded-full bg-indigo-300/20 dark:bg-blue-600/10 blur-[100px]"></div>
-      <div class="absolute inset-0 opacity-[0.04]" style="background-image: linear-gradient(rgba(255,255,255,0.6) 1px, transparent 1px), linear-gradient(90deg, rgba(255,255,255,0.6) 1px, transparent 1px); background-size: 40px 40px;"></div>
-    </div>
-
-    <!-- CTA -->
-    <div class="relative z-10 flex-1 flex flex-col items-center justify-center text-center">
-      <h2 class="text-4xl font-extrabold mb-6 text-white">即刻开启高效学习模式</h2>
-      <p class="text-blue-100/80 text-lg mb-10 max-w-xl mx-auto">告别低效抄录，让 AI 成为你的超级助教。注册即享完整试卷解析体验。</p>
-      <RouterLink to="/auth" class="relative inline-flex group h-12">
-        <div class="absolute -inset-px bg-gradient-to-r from-blue-400 via-indigo-400 to-blue-500 rounded-full blur-lg opacity-0 group-hover:opacity-80 transition-opacity duration-500"></div>
-        <span class="relative inline-flex items-center px-8 py-3 text-sm font-bold text-white bg-white/15 hover:bg-white/25 border border-white/30 rounded-full backdrop-blur-md transition-all">
-          免费创建错题库
-        </span>
-      </RouterLink>
-    </div>
-
-    <!-- Footer -->
-    <div class="relative z-10 border-t border-white/10 pt-6">
-      <div class="flex flex-col md:flex-row justify-between items-center gap-4 text-sm text-blue-100/40">
-        <div class="flex items-center gap-2">
-          <div class="bg-white/15 border border-white/20 p-1.5 rounded-xl">
-            <img src="/logo.svg" class="w-4 h-4 brightness-0 invert" alt="logo" />
-          </div>
-          <span class="font-bold text-white/60">智卷错题本</span>
-        </div>
-        <p>© 2026 Intelligent Error Book Generation System. All rights reserved.</p>
-        <div class="flex gap-6 font-medium">
-          <a href="#" class="hover:text-white transition-colors">架构文档</a>
-          <a href="#" class="hover:text-white transition-colors">隐私政策</a>
-          <a href="#" class="hover:text-white transition-colors">联系我们</a>
-        </div>
-      </div>
-    </div>
-  </section>
-</template>
diff --git a/frontend/src/components/landing/LandingDemo.vue b/frontend/src/components/landing/LandingDemo.vue
deleted file mode 100644
index 7cf49fb9..00000000
--- a/frontend/src/components/landing/LandingDemo.vue
+++ /dev/null
@@ -1,73 +0,0 @@
-<script setup>
-import { ArrowRight, Sparkles, CheckCircle2, FileDown } from 'lucide-vue-next'
-</script>
-
-<template>
-  <!-- 效果对比区 -->
-  <section id="demo" class="relative z-10 min-h-screen flex flex-col justify-center py-24 overflow-hidden">
-    <!-- 顶部过渡遮罩 -->
-    <div class="pointer-events-none absolute inset-x-0 top-0 h-32 z-10 bg-gradient-to-b from-slate-950 to-transparent"></div>
-    <!-- 装饰光晕 -->
-    <div class="pointer-events-none absolute inset-0 z-0">
-      <div class="absolute bottom-[-10%] right-[-5%] h-[500px] w-[500px] rounded-full bg-blue-400/20 dark:bg-indigo-500/10 blur-[130px]"></div>
-      <div class="absolute inset-0 opacity-[0.04]" style="background-image: linear-gradient(rgba(255,255,255,0.6) 1px, transparent 1px), linear-gradient(90deg, rgba(255,255,255,0.6) 1px, transparent 1px); background-size: 40px 40px;"></div>
-    </div>
-    <div class="reveal relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
-      <div class="text-center mb-16">
-        <h2 class="text-3xl md:text-4xl font-bold mb-6 text-white">眼见为实的 <span class="text-transparent bg-clip-text bg-gradient-to-r from-indigo-300 to-cyan-300">魔法转变</span></h2>
-        <p class="text-slate-400 text-base">复杂的公式、凌乱的涂改，在强大的 AI 引擎面前都不值一提。</p>
-      </div>
-
-      <div class="flex flex-col lg:flex-row gap-6 items-center justify-center">
-        <!-- 左侧：原图模拟 -->
-        <div class="reveal flex-1 w-full max-w-md bg-white/8 backdrop-blur-md border border-white/12 p-6 md:p-10 rounded-3xl relative overflow-hidden group">
-          <div class="absolute top-5 left-6 text-xs font-bold tracking-widest text-slate-400 uppercase">Input / 原生图像</div>
-          <div class="bg-white dark:bg-[#EAEAEA] p-8 rounded-xl transform -rotate-2 opacity-90 dark:opacity-80 mt-8 blur-[0.5px] text-slate-800 shadow-sm dark:shadow-inner filter dark:sepia-[0.2] relative">
-            <div class="hidden dark:block absolute inset-0 bg-[url(&quot;data:image/svg+xml,%3Csvg viewBox='0 0 256 256' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.7' numOctaves='4' stitchTiles='stitch'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E&quot;)] opacity-40 mix-blend-multiply"></div>
-            <p class="font-[cursive] text-base mb-3">3. 巳知函数f(x)=2sin(wx+φ)(w&gt;0,|φ|&lt;π/2)的图像...</p>
-            <p class="font-[cursive] text-base">A. w=2, φ=π/6</p>
-            <div class="mt-4 w-1/2 h-4 bg-red-400/30 dark:bg-red-400/20 rounded-full blur-[2px] transform -rotate-3"></div>
-            <p class="font-[cursive] text-sm mt-4 text-red-500 dark:text-red-600/60 font-bold transform -rotate-6">看不清，选C吧</p>
-          </div>
-        </div>
-
-        <!-- 中间发光箭头 -->
-        <div class="hidden lg:flex items-center justify-center relative w-20">
-          <div class="absolute inset-0 bg-blue-500/10 dark:bg-indigo-500/20 blur-xl rounded-full animate-pulse"></div>
-          <ArrowRight class="w-10 h-10 text-blue-500 dark:text-indigo-400 relative z-10" />
-        </div>
-        <div class="lg:hidden flex justify-center py-4">
-          <ArrowRight class="w-8 h-8 text-blue-500 dark:text-indigo-400 rotate-90" />
-        </div>
-
-        <!-- 右侧：AI 结果 -->
-        <div class="reveal flex-1 w-full max-w-md bg-white/8 backdrop-blur-md border border-indigo-400/25 p-6 md:p-10 rounded-3xl relative overflow-hidden text-left group transition-shadow duration-500" style="transition-delay: 200ms;">
-          <div class="absolute top-0 left-0 w-full h-1 bg-gradient-to-r from-blue-500 via-indigo-500 to-indigo-500 dark:from-cyan-500 dark:via-indigo-500 dark:to-indigo-500"></div>
-          <div class="absolute top-5 left-6 text-xs font-bold tracking-widest text-blue-400 dark:text-indigo-400 uppercase flex items-center gap-2">
-            <Sparkles class="w-3 h-3" /> Output / 结构化数据
-          </div>
-
-          <div class="mt-10 bg-black/20 border border-white/10 p-6 rounded-xl text-slate-200 font-mono text-sm leading-relaxed relative">
-            <div class="flex gap-2 mb-5">
-              <span class="bg-blue-500/20 dark:bg-cyan-500/10 text-blue-400 dark:text-cyan-400 px-3 py-1 rounded-full text-xs border border-blue-500/30 dark:border-cyan-500/20">选择题</span>
-              <span class="bg-indigo-500/20 dark:bg-indigo-500/10 text-indigo-400 px-3 py-1 rounded-full text-xs border border-indigo-500/30 dark:border-indigo-500/20">三角函数</span>
-            </div>
-            <p class="text-base"><span class="text-blue-500 dark:text-indigo-500 mr-2">## 3.</span> 已知函数 <span class="text-indigo-300 dark:text-indigo-300 bg-indigo-500/20 dark:bg-indigo-500/10 px-1 rounded">$f(x) = 2\sin(\omega x + \varphi)$</span> <span class="text-indigo-300 dark:text-indigo-300 bg-indigo-500/20 dark:bg-indigo-500/10 px-1 rounded">$(\omega &gt; 0, |\varphi| &lt; \frac{\pi}{2})$</span> 的图像...</p>
-            <div class="mt-5 pl-4 border-l-2 border-blue-500/50 dark:border-indigo-500/50 space-y-3">
-              <p class="flex gap-3"><span class="text-slate-400 dark:text-slate-500">A.</span> <span class="text-slate-200 dark:text-slate-300">$\omega = 2, \varphi = \frac{\pi}{6}$</span></p>
-              <p class="flex gap-3"><span class="text-slate-400 dark:text-slate-500">B.</span> <span class="text-slate-200 dark:text-slate-300">$\omega = 2, \varphi = -\frac{\pi}{6}$</span></p>
-            </div>
-            <div class="mt-8 flex items-center justify-between border-t border-slate-700 dark:border-white/10 pt-5">
-              <span class="text-xs text-emerald-400 flex items-center gap-1.5 bg-emerald-400/10 px-2 py-1 rounded">
-                <CheckCircle2 class="w-3 h-3" /> 已过滤手写痕迹
-              </span>
-              <button class="text-xs bg-white text-slate-900 hover:bg-slate-200 dark:bg-indigo-500 dark:hover:bg-indigo-400 dark:text-white px-4 py-2 rounded-lg flex items-center gap-2">
-                导出 MD <FileDown class="w-3 h-3" />
-              </button>
-            </div>
-          </div>
-        </div>
-      </div>
-    </div>
-  </section>
-</template>
diff --git a/frontend/src/components/landing/LandingFeatures.vue b/frontend/src/components/landing/LandingFeatures.vue
deleted file mode 100644
index 8c60dbb0..00000000
--- a/frontend/src/components/landing/LandingFeatures.vue
+++ /dev/null
@@ -1,58 +0,0 @@
-<script setup>
-import { Camera, Cpu, Database, BrainCircuit, Sparkles, Layers } from 'lucide-vue-next'
-
-const FEATURES = [
-  { icon: 'camera',        title: '极速录入，告别手抄',       desc: '支持 PDF/多图并行处理。单文件最高 50MB，将几十页试卷秒变结构化数据，把时间留给思考。', delay: 0 },
-  { icon: 'cpu',           title: '双引擎 OCR 极致精准',      desc: 'PaddleOCR + DeepSeek/文心一言 双重保障。自动纠错并完美还原复杂数学公式及变量下标。', delay: 80 },
-  { icon: 'database',      title: '智能图谱与一键导出',       desc: 'AI 自动提取知识点标签。自由剔除无用题目，一键导出排版精美的 Markdown / PDF 错题本。', delay: 160 },
-  { icon: 'brain-circuit', title: 'LangGraph Agent 拆题',    desc: '基于 LangGraph 的多步骤 Agent 流水线，智能识别题目边界、拆分子问题，精度超越单次 LLM 调用。', delay: 240 },
-  { icon: 'sparkles',      title: 'LaTeX 公式无损还原',       desc: '深度理解数学符号语义，自动将手写或印刷公式转换为标准 LaTeX，支持分式、积分、矩阵等复杂结构。', delay: 320 },
-  { icon: 'layers',        title: '多科目知识点打标',         desc: '覆盖数学、物理、化学、英语等主流学科。每道题自动关联知识点树，构建可视化个人薄弱点图谱。', delay: 400 },
-]
-
-const iconMap = {
-  camera: Camera,
-  cpu: Cpu,
-  database: Database,
-  'brain-circuit': BrainCircuit,
-  sparkles: Sparkles,
-  layers: Layers,
-}
-</script>
-
-<template>
-  <!-- 核心功能区 -->
-  <section id="features" class="relative overflow-hidden min-h-screen flex flex-col justify-center py-24 rounded-t-3xl bg-gradient-to-br from-blue-600 via-indigo-600 to-indigo-700 dark:from-indigo-950 dark:via-[#0A0A0F] dark:to-slate-950 shadow-[0_-12px_60px_rgba(0,0,0,0.35)]">
-    <!-- 装饰光晕 -->
-    <div class="pointer-events-none absolute inset-0 z-0">
-      <div class="absolute top-[-15%] left-[-8%] h-[500px] w-[500px] rounded-full bg-blue-400/20 dark:bg-indigo-500/10 blur-[130px]"></div>
-      <div class="absolute bottom-[-10%] right-[-5%] h-[400px] w-[400px] rounded-full bg-indigo-300/20 dark:bg-blue-600/10 blur-[100px]"></div>
-      <div class="absolute inset-0 opacity-[0.04]" style="background-image: linear-gradient(rgba(255,255,255,0.6) 1px, transparent 1px), linear-gradient(90deg, rgba(255,255,255,0.6) 1px, transparent 1px); background-size: 40px 40px;"></div>
-    </div>
-
-    <div class="relative z-10 mx-auto max-w-7xl px-4 sm:px-6 lg:px-8">
-      <div class="mb-12 text-center">
-        <h2 class="reveal mb-6 text-3xl font-bold text-white">驱动学习效率的 <span class="text-blue-200 dark:text-indigo-300">核心引擎</span></h2>
-        <p class="reveal mx-auto max-w-2xl text-base text-blue-100/80 dark:text-slate-400">不仅是简单的图像识别，而是真正理解学科内在逻辑的 AI 智能体系统。</p>
-      </div>
-      <div class="grid gap-6 md:grid-cols-3">
-        <div
-          v-for="f in FEATURES"
-          :key="f.title"
-          class="reveal group relative rounded-3xl border border-white/15 bg-white/10 backdrop-blur-md p-8 hover:-translate-y-2 hover:bg-white/15 hover:border-white/25 transition-all duration-300"
-          :style="f.delay ? `transition-delay:${f.delay}ms` : ''"
-        >
-          <div class="relative z-10">
-            <div class="mb-6 flex h-14 w-14 items-center justify-center rounded-xl border border-white/20 bg-white/15 text-white shadow-sm group-hover:scale-110 transition-transform duration-300">
-              <component :is="iconMap[f.icon]" class="h-7 w-7" />
-            </div>
-            <h3 class="mb-4 text-xl font-bold text-white">{{ f.title }}</h3>
-            <p class="text-sm leading-relaxed text-blue-100/80 dark:text-slate-300">{{ f.desc }}</p>
-          </div>
-        </div>
-      </div>
-    </div>
-    <!-- 底部过渡遮罩 -->
-    <div class="pointer-events-none absolute inset-x-0 bottom-0 h-32 z-10 bg-gradient-to-t from-slate-50 to-transparent dark:from-[#0A0A0F] dark:to-transparent"></div>
-  </section>
-</template>
diff --git a/frontend/src/components/landing/LandingHero.vue b/frontend/src/components/landing/LandingHero.vue
deleted file mode 100644
index 5467ae47..00000000
--- a/frontend/src/components/landing/LandingHero.vue
+++ /dev/null
@@ -1,347 +0,0 @@
-<script setup>
-import { ref, onMounted, onUnmounted } from 'vue'
-import { Zap, UploadCloud, ArrowRight, Terminal, CheckCircle2, Sparkles } from 'lucide-vue-next'
-
-const canvasRef = ref(null)
-
-let raf = null
-let resizeUnlisten = null
-
-// ── Canvas neural animation ──
-function initCanvas(canvas) {
-  const ctx = canvas.getContext('2d')
-  const NODE_COUNT = 72
-  const CONNECT_DIST = 180
-  const MAX_CONNECTIONS = 4
-  let W, H, nodes, edges, pulses
-
-  function isDark() { return document.documentElement.classList.contains('dark') }
-
-  function resize() {
-    W = canvas.width = canvas.offsetWidth
-    H = canvas.height = canvas.offsetHeight
-    init()
-  }
-
-  function init() {
-    nodes = Array.from({ length: NODE_COUNT }, () => ({
-      x: Math.random() * W,
-      y: Math.random() * H,
-      vx: (Math.random() - 0.5) * 0.3,
-      vy: (Math.random() - 0.5) * 0.3,
-      r: 2 + Math.random() * 2,
-      activation: 0,
-    }))
-    pulses = []
-    for (let i = 0; i < 18; i++) spawnPulse()
-  }
-
-  function buildEdges() {
-    edges = []
-    const used = new Array(NODE_COUNT).fill(0)
-    for (let i = 0; i < NODE_COUNT; i++) {
-      const neighbors = []
-      for (let j = 0; j < NODE_COUNT; j++) {
-        if (i === j) continue
-        const dx = nodes[i].x - nodes[j].x
-        const dy = nodes[i].y - nodes[j].y
-        const d = Math.sqrt(dx * dx + dy * dy)
-        if (d < CONNECT_DIST) neighbors.push({ j, d })
-      }
-      neighbors.sort((a, b) => a.d - b.d)
-      neighbors.slice(0, MAX_CONNECTIONS).forEach(({ j, d }) => {
-        if (used[i] < MAX_CONNECTIONS && used[j] < MAX_CONNECTIONS) {
-          if (!edges.find(e => (e.a === i && e.b === j) || (e.a === j && e.b === i))) {
-            edges.push({ a: i, b: j, alpha: 1 - d / CONNECT_DIST })
-            used[i]++
-            used[j]++
-          }
-        }
-      })
-    }
-  }
-
-  function spawnPulse() {
-    if (!edges || edges.length === 0) return
-    const edge = edges[Math.floor(Math.random() * edges.length)]
-    pulses.push({
-      edge,
-      progress: 0,
-      speed: 0.012 + Math.random() * 0.012,
-    })
-  }
-
-  let lastSpawn = 0
-  function update(ts) {
-    nodes.forEach(n => {
-      n.x += n.vx
-      n.y += n.vy
-      if (n.x < 0 || n.x > W) n.vx *= -1
-      if (n.y < 0 || n.y > H) n.vy *= -1
-      n.activation *= 0.94
-    })
-
-    buildEdges()
-
-    for (let i = pulses.length - 1; i >= 0; i--) {
-      const p = pulses[i]
-      p.progress += p.speed
-      if (p.progress >= 1) {
-        const targetIdx = p.edge.b
-        nodes[targetIdx].activation = 1
-        pulses.splice(i, 1)
-        const outEdges = edges.filter(e => e.a === targetIdx || e.b === targetIdx)
-        if (outEdges.length && Math.random() < 0.7) {
-          const next = outEdges[Math.floor(Math.random() * outEdges.length)]
-          const fwd = next.a === targetIdx ? next : { a: next.b, b: next.a, alpha: next.alpha }
-          pulses.push({ edge: fwd, progress: 0, speed: 0.012 + Math.random() * 0.012 })
-        }
-      }
-    }
-
-    if (ts - lastSpawn > 800) {
-      spawnPulse()
-      lastSpawn = ts
-    }
-  }
-
-  function draw() {
-    ctx.clearRect(0, 0, W, H)
-    const dark = isDark()
-
-    edges.forEach(({ a, b, alpha }) => {
-      const na = nodes[a], nb = nodes[b]
-      const act = Math.max(na.activation, nb.activation)
-      const baseAlpha = dark ? 0.08 : 0.06
-      const lineAlpha = baseAlpha + act * (dark ? 0.25 : 0.2)
-      ctx.beginPath()
-      ctx.moveTo(na.x, na.y)
-      ctx.lineTo(nb.x, nb.y)
-      ctx.strokeStyle = dark
-        ? `rgba(99,102,241,${(lineAlpha * alpha).toFixed(3)})`
-        : `rgba(59,130,246,${(lineAlpha * alpha).toFixed(3)})`
-      ctx.lineWidth = 0.8 + act * 0.6
-      ctx.stroke()
-    })
-
-    nodes.forEach(n => {
-      const act = n.activation
-      const baseR = n.r
-      const r = baseR + act * 4
-
-      if (act > 0.05) {
-        const grd = ctx.createRadialGradient(n.x, n.y, 0, n.x, n.y, r * 4)
-        grd.addColorStop(0, dark ? `rgba(165,180,252,${act * 0.5})` : `rgba(59,130,246,${act * 0.4})`)
-        grd.addColorStop(1, 'transparent')
-        ctx.beginPath()
-        ctx.arc(n.x, n.y, r * 4, 0, Math.PI * 2)
-        ctx.fillStyle = grd
-        ctx.fill()
-      }
-
-      ctx.beginPath()
-      ctx.arc(n.x, n.y, r, 0, Math.PI * 2)
-      ctx.fillStyle = dark
-        ? `rgba(${act > 0.1 ? '165,180,252' : '99,102,241'},${0.2 + act * 0.7})`
-        : `rgba(${act > 0.1 ? '59,130,246' : '148,163,184'},${0.25 + act * 0.6})`
-      ctx.fill()
-    })
-
-    pulses.forEach(p => {
-      const na = nodes[p.edge.a], nb = nodes[p.edge.b]
-      const x = na.x + (nb.x - na.x) * p.progress
-      const y = na.y + (nb.y - na.y) * p.progress
-
-      const grd2 = ctx.createRadialGradient(x, y, 0, x, y, 22)
-      grd2.addColorStop(0, dark ? 'rgba(99,102,241,0.45)' : 'rgba(59,130,246,0.35)')
-      grd2.addColorStop(1, 'transparent')
-      ctx.beginPath()
-      ctx.arc(x, y, 22, 0, Math.PI * 2)
-      ctx.fillStyle = grd2
-      ctx.fill()
-
-      const grd = ctx.createRadialGradient(x, y, 0, x, y, 10)
-      grd.addColorStop(0, dark ? 'rgba(224,231,255,1)' : 'rgba(59,130,246,0.95)')
-      grd.addColorStop(0.35, dark ? 'rgba(99,102,241,0.7)' : 'rgba(99,102,241,0.55)')
-      grd.addColorStop(1, 'transparent')
-      ctx.beginPath()
-      ctx.arc(x, y, 10, 0, Math.PI * 2)
-      ctx.fillStyle = grd
-      ctx.fill()
-
-      ctx.beginPath()
-      ctx.arc(x, y, 2.5, 0, Math.PI * 2)
-      ctx.fillStyle = dark ? 'rgba(224,231,255,1)' : 'rgba(59,130,246,1)'
-      ctx.fill()
-    })
-  }
-
-  function loop(ts) {
-    update(ts)
-    draw()
-    raf = requestAnimationFrame(loop)
-  }
-
-  const onResize = () => resize()
-  window.addEventListener('resize', onResize)
-  resizeUnlisten = () => window.removeEventListener('resize', onResize)
-
-  resize()
-  raf = requestAnimationFrame(loop)
-  requestAnimationFrame(() => { canvas.style.opacity = '1' })
-}
-
-onMounted(() => {
-  if (canvasRef.value) {
-    initCanvas(canvasRef.value)
-  }
-})
-
-onUnmounted(() => {
-  if (resizeUnlisten) resizeUnlisten()
-  if (raf) cancelAnimationFrame(raf)
-})
-</script>
-
-<template>
-  <!-- ① Sticky Hero 容器 -->
-  <div id="sticky-hero" class="sticky top-0 h-screen overflow-hidden z-0 flex items-center bg-slate-50 dark:bg-[#0A0A0F]">
-
-    <!-- 动态背景环境光 -->
-    <div class="absolute inset-0 overflow-hidden pointer-events-none z-0 h-screen">
-      <div class="absolute top-[-10%] left-[-10%] w-[40vw] h-[40vw] rounded-full bg-blue-300/20 dark:bg-indigo-600/20 dark:mix-blend-screen filter blur-[100px] animate-blob"></div>
-      <div class="absolute top-[20%] right-[-10%] w-[35vw] h-[35vw] rounded-full bg-indigo-200/25 dark:bg-indigo-600/20 dark:mix-blend-screen filter blur-[100px] animate-blob animation-delay-2000"></div>
-      <div class="absolute bottom-[-20%] left-[20%] w-[50vw] h-[50vw] rounded-full bg-cyan-200/20 dark:bg-cyan-600/10 dark:mix-blend-screen filter blur-[120px] animate-blob animation-delay-4000"></div>
-      <div class="absolute inset-0 bg-[url(&quot;data:image/svg+xml,%3Csvg viewBox='0 0 256 256' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.7' numOctaves='4' stitchTiles='stitch'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E&quot;)] opacity-[0.03] dark:opacity-20 mix-blend-overlay"></div>
-    </div>
-
-    <!-- 电路板背景动画 -->
-    <canvas
-      ref="canvasRef"
-      id="circuit-canvas"
-      class="absolute inset-0 pointer-events-none z-0"
-      style="height:100vh;width:100%;opacity:0;transition:opacity 1.2s ease;"
-    ></canvas>
-
-    <!-- 首屏区块 -->
-    <section id="hero" class="relative w-full pt-20 pb-8 px-4 sm:px-6 lg:px-8 max-w-7xl mx-auto flex flex-col lg:flex-row items-center gap-16 z-10" style="transform-origin: center 55%; will-change: transform;">
-      <div class="flex-1 text-center lg:text-left">
-        <div class="hero-anim">
-          <div class="mb-10 flex items-center justify-center lg:justify-start">
-            <div class="relative flex items-center gap-3 py-1 text-xs font-black tracking-widest text-blue-700/90 dark:text-indigo-300/90">
-              <div class="relative flex h-5 w-5 items-center justify-center">
-                <Zap class="absolute w-4 h-4 text-orange-500 dark:text-yellow-400 animate-pulse" />
-                <div class="absolute h-full w-full animate-ping rounded-full bg-orange-400/10 dark:bg-yellow-400/10"></div>
-              </div>
-              <span class="relative z-10 uppercase pb-0.5">
-                新一代 AI 错题处理架构 <span class="ml-1 font-extrabold text-blue-600 dark:text-indigo-300">V2.0</span>
-              </span>
-            </div>
-          </div>
-          <h1 class="text-4xl font-extrabold tracking-tight leading-[1.1] mb-6 text-slate-900 dark:text-white">
-            重塑错题整理 <br />
-            <span class="text-transparent bg-clip-text bg-gradient-to-r from-blue-600 to-indigo-600 dark:from-cyan-400 dark:via-indigo-400 dark:to-indigo-400 animate-pulse">
-              一键生成知识图谱
-            </span>
-          </h1>
-          <p class="text-base md:text-xl text-slate-600 dark:text-slate-400 mb-10 max-w-2xl mx-auto lg:mx-0 leading-relaxed">
-            专为中学生与大学生研发。上传凌乱试卷，AI 自动完成图片分割、OCR 纠错及 LaTeX 公式还原。释放你的双手，将时间交还给真正的思考。
-          </p>
-          <div class="flex flex-col sm:flex-row gap-4 justify-center lg:justify-start" style="transition-delay: 450ms;">
-            <RouterLink to="/auth" class="relative inline-flex group h-14 w-full sm:w-auto">
-              <div class="absolute -inset-px bg-gradient-to-r from-blue-400 via-indigo-400 to-blue-500 rounded-full blur-lg opacity-0 group-hover:opacity-80 transition-opacity duration-500"></div>
-              <span class="relative inline-flex items-center justify-center w-full px-8 py-4 text-base font-bold text-white bg-white/15 hover:bg-white/25 border border-white/30 rounded-full backdrop-blur-md transition-all gap-3">
-                <UploadCloud class="w-5 h-5" />
-                上传试卷分析
-              </span>
-            </RouterLink>
-            <a href="#demo" class="relative inline-flex group h-14 w-full sm:w-auto">
-              <div class="absolute -inset-px bg-gradient-to-r from-blue-400 via-indigo-400 to-blue-500 rounded-full blur-lg opacity-0 group-hover:opacity-80 transition-opacity duration-500"></div>
-              <span class="relative inline-flex items-center justify-center w-full px-8 py-4 text-base font-bold text-white bg-white/15 hover:bg-white/25 border border-white/30 rounded-full backdrop-blur-md transition-all gap-2">
-                查看实时演示
-                <ArrowRight class="w-5 h-5" />
-              </span>
-            </a>
-          </div>
-        </div>
-      </div>
-
-      <!-- 右侧代码状态图 -->
-      <div class="hero-anim flex-1 relative w-full max-w-xl lg:max-w-none" style="transition-delay: 200ms;">
-        <div class="relative animate-float group">
-
-          <!-- 后层：空窗口 -->
-          <div class="absolute inset-0 translate-x-6 -translate-y-5 z-0 opacity-80 dark:opacity-60 pointer-events-none transition-transform duration-500 ease-out group-hover:translate-x-10 group-hover:-translate-y-8">
-            <div class="relative bg-white dark:bg-transparent dark:glass-panel rounded-2xl p-1 border border-slate-200 dark:border-indigo-500/40 shadow-md dark:shadow-[0_0_20px_rgba(99,102,241,0.2)] h-full">
-              <div class="bg-slate-50 dark:bg-[#0A0A0F]/80 backdrop-blur-md rounded-t-xl p-4 flex items-center justify-between border-b border-slate-100 dark:border-white/5">
-                <div class="flex gap-2">
-                  <div class="w-3 h-3 rounded-full bg-red-400 dark:bg-[#FF5F57]"></div>
-                  <div class="w-3 h-3 rounded-full bg-yellow-400 dark:bg-[#FFBD2E]"></div>
-                  <div class="w-3 h-3 rounded-full bg-green-400 dark:bg-[#28C840]"></div>
-                </div>
-                <div class="flex items-center gap-2 text-xs font-mono text-slate-400 dark:text-slate-600">
-                  <Terminal class="w-3 h-3" />
-                  error_bank.py
-                </div>
-              </div>
-              <div class="bg-white dark:bg-[#0A0A0F]/90 rounded-b-xl h-[340px]"></div>
-            </div>
-          </div>
-
-          <!-- 前层：有内容的卡片 -->
-          <div class="relative overflow-hidden rounded-2xl z-10 transition-transform duration-500 ease-out group-hover:-translate-x-2 group-hover:translate-y-1">
-            <div class="absolute inset-0 bg-gradient-to-tr from-blue-200 to-indigo-200 dark:from-indigo-500 dark:to-indigo-500 rounded-3xl transform rotate-6 scale-105 opacity-60 dark:opacity-20 blur-xl"></div>
-            <div class="absolute inset-0 bg-gradient-to-bl from-cyan-200 to-blue-200 dark:from-cyan-500 dark:to-blue-500 rounded-3xl transform -rotate-3 scale-105 opacity-60 dark:opacity-20 blur-xl"></div>
-            <div class="bg-white dark:bg-transparent dark:glass-panel rounded-2xl p-1 relative overflow-hidden border border-slate-200 dark:border-indigo-500/40 shadow-md dark:shadow-[0_0_20px_rgba(99,102,241,0.2)]">
-              <div class="bg-slate-50 dark:bg-[#0A0A0F]/80 backdrop-blur-md rounded-t-xl p-4 flex items-center justify-between border-b border-slate-100 dark:border-white/5">
-                <div class="flex gap-2">
-                  <div class="w-3 h-3 rounded-full bg-red-400 dark:bg-[#FF5F57]"></div>
-                  <div class="w-3 h-3 rounded-full bg-yellow-400 dark:bg-[#FFBD2E]"></div>
-                  <div class="w-3 h-3 rounded-full bg-green-400 dark:bg-[#28C840]"></div>
-                </div>
-                <div class="flex items-center gap-2 text-xs font-mono text-slate-500">
-                  <Terminal class="w-3 h-3" />
-                  agent_workflow.py
-                </div>
-              </div>
-
-              <div class="bg-white dark:bg-[#0A0A0F]/90 p-6 rounded-b-xl font-mono text-sm space-y-5 h-[340px] overflow-hidden relative">
-                <div class="flex items-start gap-3">
-                  <span class="text-blue-500 dark:text-cyan-500 mt-0.5">❯</span>
-                  <div>
-                    <span class="text-slate-700 dark:text-slate-300">init </span>
-                    <span class="text-indigo-600 dark:text-indigo-400 font-bold">PaddleOCR_v2 </span>
-                    <span class="text-slate-400 dark:text-slate-500">--async</span>
-                  </div>
-                </div>
-                <div class="flex items-start gap-3 text-emerald-600 dark:text-emerald-400">
-                  <CheckCircle2 class="w-4 h-4 mt-0.5" />
-                  <span>[OCR] 解析完成 (0.8s) - 提取 2,451 字符</span>
-                </div>
-
-                <div class="flex items-start gap-3 relative">
-                  <div class="absolute -left-2 top-0 w-0.5 h-full bg-indigo-500 shadow-indigo-500/50"></div>
-                  <Sparkles class="w-4 h-4 mt-0.5 text-indigo-500 dark:text-indigo-400" />
-                  <div class="flex-1">
-                    <p class="text-indigo-600 dark:text-indigo-300 mb-2 font-semibold">LangGraph.invoke(solve_agent)...</p>
-                    <div class="bg-slate-50 dark:bg-[#0A0A0F] border border-slate-200 dark:border-white/5 rounded-lg p-3 text-xs text-slate-500 dark:text-slate-400">
-                      <p class="text-indigo-600 dark:text-indigo-400 mb-1"># Fixing OCR artifacts &amp; converting to LaTeX</p>
-                      <p>original: <span class="text-red-500/80 dark:text-red-400/70 line-through">f(x)=sin(wx+φ)</span></p>
-                      <p>fixed: &nbsp;&nbsp;<span class="text-emerald-600 dark:text-green-400">$f(x) = \sin(\omega x + \varphi)$</span></p>
-                    </div>
-                  </div>
-                </div>
-
-                <div class="flex items-center gap-3 text-slate-400 dark:text-slate-500 animate-pulse">
-                  <span class="text-blue-500 dark:text-cyan-500">❯</span>
-                  Generating Markdown structured output_
-                </div>
-              </div>
-            </div>
-          </div><!-- /光晕裁切层 -->
-        </div>
-      </div>
-    </section>
-
-  </div><!-- /sticky-hero -->
-</template>
diff --git a/frontend/src/components/landing/LandingNav.vue b/frontend/src/components/landing/LandingNav.vue
deleted file mode 100644
index 8f3dff27..00000000
--- a/frontend/src/components/landing/LandingNav.vue
+++ /dev/null
@@ -1,90 +0,0 @@
-<script setup>
-import { ArrowRight } from 'lucide-vue-next'
-
-const props = defineProps({
-  navScrolled: Boolean,
-  activeSection: String,
-  indicatorTop: Number,
-  sections: Array,
-})
-
-const emit = defineEmits(['scrollToSection'])
-</script>
-
-<template>
-  <!-- 导航栏 -->
-  <nav
-    id="top-nav"
-    class="fixed top-0 left-0 w-full z-50 transition-all duration-300 border-b border-transparent"
-    :class="navScrolled ? 'glass-panel !border-white/5' : 'border-transparent'"
-  >
-    <div class="max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
-      <div
-        id="nav-container"
-        class="flex justify-between items-center transition-all duration-300"
-        :class="navScrolled ? 'h-16' : 'h-20'"
-      >
-        <div class="flex items-center gap-3">
-          <div class="relative group">
-            <div class="absolute inset-0 bg-blue-500/60 dark:bg-indigo-500/60 blur-xl opacity-0 group-hover:opacity-100 transition duration-500 rounded-2xl"></div>
-            <div class="relative bg-gradient-to-br from-blue-500 to-indigo-700 dark:from-indigo-500 dark:to-indigo-800 p-2.5 rounded-2xl shadow-md shadow-indigo-500/30 border border-white/10">
-              <img src="/logo.svg" class="w-7 h-7 brightness-0 invert" alt="logo" />
-            </div>
-          </div>
-          <span class="text-2xl font-bold bg-clip-text text-transparent bg-gradient-to-r from-blue-700 to-indigo-700 dark:from-white dark:via-indigo-200 dark:to-indigo-200 tracking-wide">
-            智卷错题本
-          </span>
-        </div>
-        <div class="hidden md:flex space-x-8">
-          <button data-section="features" class="text-slate-600 hover:text-blue-600 dark:text-slate-400 dark:hover:text-white text-sm font-semibold tracking-wider" @click="emit('scrollToSection', 'features')">核心功能</button>
-          <button data-section="workflow" class="text-slate-600 hover:text-blue-600 dark:text-slate-400 dark:hover:text-white text-sm font-semibold tracking-wider" @click="emit('scrollToSection', 'workflow')">工作流</button>
-          <button data-section="demo" class="text-slate-600 hover:text-blue-600 dark:text-slate-400 dark:hover:text-white text-sm font-semibold tracking-wider" @click="emit('scrollToSection', 'demo')">效果演示</button>
-          <button data-section="cta" class="text-slate-600 hover:text-blue-600 dark:text-slate-400 dark:hover:text-white text-sm font-semibold tracking-wider" @click="emit('scrollToSection', 'cta')">立即开始</button>
-        </div>
-
-        <div class="flex items-center gap-4">
-          <a href="https://github.com/xiaozhejiya/error_correction" target="_blank" rel="noopener noreferrer" class="text-slate-400 hover:text-white transition-colors">
-            <i class="fa-brands fa-github text-xl"></i>
-          </a>
-          <RouterLink to="/auth" class="relative inline-flex group h-12 active:scale-95 transition-transform">
-            <div class="absolute -inset-px bg-gradient-to-r from-blue-400 via-indigo-400 to-blue-500 rounded-full blur-lg opacity-0 group-hover:opacity-80 transition-opacity duration-500"></div>
-            <span class="relative inline-flex items-center justify-center px-6 py-2 text-sm font-bold text-white bg-white/15 hover:bg-white/25 border border-white/30 rounded-full backdrop-blur-md transition-all">
-              进入工作台
-            </span>
-          </RouterLink>
-        </div>
-      </div>
-    </div>
-  </nav>
-
-  <!-- 右侧 Section 导航 -->
-  <nav id="section-nav" class="fixed right-4 md:right-8 top-1/2 -translate-y-1/2 z-40 flex flex-col items-center py-6 px-2 rounded-full border border-slate-200/60 bg-white/60 backdrop-blur-md dark:border-white/5 dark:bg-white/[0.02] scale-90 md:scale-100">
-    <!-- 垂直轨道 -->
-    <div class="absolute top-8 bottom-8 w-px bg-slate-200 dark:bg-white/10 pointer-events-none"></div>
-    <!-- 激活指示器 -->
-    <div
-      id="section-nav-indicator"
-      class="absolute w-2 h-2 bg-blue-600 dark:bg-indigo-400 rounded-full pointer-events-none z-10 transition-all duration-500 shadow-blue-600/70 dark:shadow-indigo-400/80"
-      :style="{ top: indicatorTop + 'px', transform: 'translateY(-50%)' }"
-    ></div>
-
-    <button
-      v-for="(s, i) in sections"
-      :key="s.id"
-      class="section-nav-btn relative flex items-center justify-center w-10 h-10 focus:outline-none group"
-      :title="s.label"
-      :data-id="s.id"
-      @click="emit('scrollToSection', s.id)"
-    >
-      <div
-        class="section-dot w-1.5 h-1.5 rounded-full transition-all duration-300"
-        :class="activeSection === s.id
-          ? 'bg-blue-600 dark:bg-indigo-400 scale-125 shadow-[0_0_8px_3px_rgba(99,102,241,0.7)] dark:shadow-[0_0_10px_4px_rgba(99,102,241,0.9)]'
-          : 'bg-slate-300 dark:bg-white/20 group-hover:bg-blue-400 dark:group-hover:bg-indigo-300'"
-      ></div>
-      <span class="absolute right-14 px-3 py-1.5 rounded-xl bg-white dark:bg-slate-900 border border-slate-200 dark:border-white/10 text-xs font-black text-slate-700 dark:text-white whitespace-nowrap pointer-events-none opacity-0 translate-x-2 group-hover:opacity-100 group-hover:translate-x-0 shadow-md transition-all duration-300">
-        <span class="opacity-40 mr-2">{{ String(i + 1).padStart(2, '0') }}</span>{{ s.label }}
-      </span>
-    </button>
-  </nav>
-</template>
diff --git a/frontend/src/components/landing/LandingWorkflow.vue b/frontend/src/components/landing/LandingWorkflow.vue
deleted file mode 100644
index 0ce7bf3d..00000000
--- a/frontend/src/components/landing/LandingWorkflow.vue
+++ /dev/null
@@ -1,122 +0,0 @@
-<script setup>
-import { ref, onMounted, onUnmounted } from 'vue'
-import { UploadCloud, BrainCircuit, Sparkles, FileDown } from 'lucide-vue-next'
-
-const STEPS = [
-  { icon: 'upload-cloud', num: '01', title: '上传试卷',   desc: '支持PDF/多图并行处理',  delay: 0 },
-  { icon: 'brain-circuit', num: '02', title: 'AI 解析拆题', desc: '智能排版与图文切分',    delay: 150 },
-  { icon: 'sparkles',     num: '03', title: '纠错与打标', desc: '公式还原与知识点关联',   delay: 300 },
-  { icon: 'file-down',    num: '04', title: '一键导出',   desc: '生成 Markdown/PDF',      delay: 450 },
-]
-
-const iconMap = {
-  'upload-cloud': UploadCloud,
-  'brain-circuit': BrainCircuit,
-  sparkles: Sparkles,
-  'file-down': FileDown,
-}
-
-const activeStep = ref(0)
-let intervalId = null
-
-function startStepInterval() {
-  intervalId = setInterval(() => {
-    activeStep.value = (activeStep.value + 1) % STEPS.length
-  }, 3000)
-}
-
-function onStepClick(idx) {
-  clearInterval(intervalId)
-  activeStep.value = idx
-  startStepInterval()
-}
-
-function getStepIconClass(idx) {
-  if (idx === activeStep.value) {
-    return 'step-icon w-16 h-16 rounded-xl flex items-center justify-center mb-6 relative bg-blue-600 text-white shadow-md shadow-blue-500/30 dark:bg-gradient-to-br dark:from-indigo-500 dark:to-indigo-600 dark:shadow-[0_0_20px_rgba(99,102,241,0.4)]'
-  } else if (idx < activeStep.value) {
-    return 'step-icon w-16 h-16 rounded-xl flex items-center justify-center mb-6 relative bg-blue-600 text-white shadow-md dark:bg-gradient-to-br dark:from-indigo-500 dark:to-indigo-600 dark:shadow-[0_0_20px_rgba(99,102,241,0.4)]'
-  }
-  return 'step-icon w-16 h-16 rounded-xl flex items-center justify-center mb-6 relative bg-slate-200 text-slate-500 border border-slate-200 dark:bg-white/5 dark:text-slate-500 dark:border-white/10'
-}
-
-function getStepContainerClass(idx) {
-  if (idx === activeStep.value) {
-    return 'workflow-step relative p-6 rounded-2xl cursor-pointer bg-white border border-blue-500 shadow-md scale-105 dark:bg-transparent dark:glass-panel dark:border-indigo-500/50 dark:shadow-[0_0_30px_rgba(99,102,241,0.15)]'
-  }
-  return 'workflow-step relative p-6 rounded-2xl cursor-pointer bg-transparent border border-transparent hover:bg-slate-100 dark:hover:bg-white/5'
-}
-
-function getStepTitleClass(idx) {
-  if (idx === activeStep.value) return 'step-title text-xl font-bold mb-2 text-slate-900 dark:text-white'
-  return 'step-title text-xl font-bold mb-2 text-slate-500 dark:text-slate-400'
-}
-
-function getStepDescClass(idx) {
-  if (idx === activeStep.value) return 'step-desc text-sm text-blue-600 dark:text-indigo-200'
-  return 'step-desc text-sm text-slate-500 dark:text-slate-500'
-}
-
-function getProgressWidth() {
-  return `${(activeStep.value / (STEPS.length - 1)) * 100}%`
-}
-
-onMounted(() => {
-  startStepInterval()
-})
-
-onUnmounted(() => {
-  if (intervalId) clearInterval(intervalId)
-})
-</script>
-
-<template>
-  <!-- 工作流演示区 -->
-  <section id="workflow" class="relative z-10 min-h-screen flex flex-col justify-center py-24 overflow-hidden bg-gradient-to-b from-slate-950 via-slate-900 to-slate-950 dark:from-[#0A0A0F] dark:via-slate-950 dark:to-[#0A0A0F]">
-    <!-- 装饰光晕 -->
-    <div class="pointer-events-none absolute inset-0 z-0">
-      <div class="absolute top-1/4 left-[-10%] w-[500px] h-[500px] rounded-full bg-indigo-600/10 blur-[140px]"></div>
-      <div class="absolute bottom-1/4 right-[-10%] w-[400px] h-[400px] rounded-full bg-cyan-600/10 blur-[120px]"></div>
-    </div>
-    <div class="reveal relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
-      <div class="text-center mb-20">
-        <h2 class="text-3xl md:text-4xl font-bold mb-6 text-white">极简四步，自动运转</h2>
-        <p class="text-slate-400 text-base">将原本需要耗费数小时的繁杂抄录，浓缩进点击之间。</p>
-      </div>
-
-      <div class="relative max-w-5xl mx-auto">
-        <div class="hidden lg:block absolute top-1/2 left-[10%] right-[10%] h-0.5 bg-white/8 -translate-y-1/2 z-0 rounded-full">
-          <div
-            id="progress-bar"
-            class="absolute top-0 left-0 h-full bg-blue-500 dark:bg-gradient-to-r dark:from-cyan-500 dark:via-indigo-500 dark:to-indigo-500 shadow-blue-500/50 dark:shadow-indigo-500/50 transition-all duration-700 ease-in-out"
-            :style="{ width: getProgressWidth() }"
-          ></div>
-        </div>
-
-        <div class="grid grid-cols-1 lg:grid-cols-4 gap-8 relative z-10">
-          <div
-            v-for="(s, i) in STEPS"
-            :key="s.num"
-            :class="getStepContainerClass(i)"
-            :data-index="i"
-            @click="onStepClick(i)"
-          >
-            <div class="flex flex-col items-center text-center">
-              <div :class="getStepIconClass(i)">
-                <div
-                  class="step-ping absolute inset-0 rounded-2xl bg-indigo-400 animate-ping opacity-20"
-                  :class="i === activeStep ? '' : 'hidden'"
-                ></div>
-                <component :is="iconMap[s.icon]" class="w-6 h-6" />
-              </div>
-              <h4 :class="getStepTitleClass(i)">
-                <span class="text-blue-500 dark:text-indigo-500 mr-2 font-mono text-sm">{{ s.num }}</span> {{ s.title }}
-              </h4>
-              <p :class="getStepDescClass(i)">{{ s.desc }}</p>
-            </div>
-          </div>
-        </div>
-      </div>
-    </div>
-  </section>
-</template>
diff --git a/frontend/src/components/EditNoteDialog.vue b/frontend/src/components/question/EditNoteDialog.vue
similarity index 97%
rename from frontend/src/components/EditNoteDialog.vue
rename to frontend/src/components/question/EditNoteDialog.vue
index 19572433..6ba3e0fb 100644
--- a/frontend/src/components/EditNoteDialog.vue
+++ b/frontend/src/components/question/EditNoteDialog.vue
@@ -1,6 +1,10 @@
 <script setup>
+/**
+ * EditNoteDialog.vue
+ * 题目/笔记编辑弹窗（Markdown 编辑）
+ */
 import { ref, watch, nextTick } from 'vue'
-import { typesetMath } from '../utils.js'
+import { typesetMath } from '@/utils.js'
 
 const props = defineProps({
   open: { type: Boolean, default: false },
@@ -69,7 +73,7 @@ const config = () => {
   <Teleport to="body">
     <Transition name="dialog-fade">
       <div v-if="open" class="fixed inset-0 z-[101] flex items-center justify-center p-4 md:left-64" @click.self="emit('close')">
-        <div class="absolute inset-0 bg-black/40 backdrop-blur-sm"></div>
+        <div class="absolute inset-0 bg-black/40"></div>
 
         <div class="relative w-full rounded-2xl border border-slate-200/60 bg-white shadow-2xl dark:border-white/10 dark:bg-[#0f0f17]"
              :class="field === 'question' ? 'max-w-2xl' : 'max-w-xl'">
diff --git a/frontend/src/components/QuestionCard.vue b/frontend/src/components/question/QuestionCard.vue
similarity index 59%
rename from frontend/src/components/QuestionCard.vue
rename to frontend/src/components/question/QuestionCard.vue
index 6d59fd3b..341c0e18 100644
--- a/frontend/src/components/QuestionCard.vue
+++ b/frontend/src/components/question/QuestionCard.vue
@@ -1,6 +1,10 @@
 <script setup>
-import { ref } from 'vue'
-import { isHtml, sanitizeHtml, formatOption } from '../utils.js'
+/**
+ * QuestionCard.vue
+ * 题目卡片（错题库列表项）
+ */
+import { ref, computed } from 'vue'
+import { isHtml, sanitizeHtml, formatOption } from '@/utils.js'
 
 const props = defineProps({
   question: { type: Object, required: true },
@@ -9,6 +13,38 @@ const props = defineProps({
 
 const emit = defineEmits(['toggle', 'open-image'])
 
+// 兜底：image_refs 中有但 content_blocks 中未嵌入的图片
+const extraImages = computed(() => {
+  const refs = props.question.image_refs || []
+  if (!refs.length) return []
+  const blocks = props.question.content_blocks || []
+  // 排除 image block 中已嵌入的
+  const embedded = new Set(
+    blocks.filter(b => b.block_type === 'image' && b.content).map(b => b.content)
+  )
+  // 排除 HTML 文本中已通过 <img> 引用的（如表格内嵌图片）
+  const htmlContent = blocks.filter(b => b.block_type === 'text').map(b => b.content || '').join(' ')
+  return refs.filter(r => {
+    if (embedded.has(r)) return false
+    // 提取文件名部分匹配，因为 HTML 中可能用 imgs/ 而 refs 用 /images/
+    const filename = r.split('/').pop()
+    if (filename && htmlContent.includes(filename)) return false
+    return true
+  })
+})
+
+// 图片与选项配对：优先使用 Agent 输出的 option_images，降级到按数量匹配
+const optionImages = computed(() => {
+  const opts = props.question.options || []
+  if (!opts.length) return null
+  // 优先使用 Agent 直接输出的 option_images
+  const agentImages = props.question.option_images
+  if (agentImages?.length === opts.length) return agentImages
+  // 降级：extraImages 数量匹配时按索引配对
+  if (extraImages.value.length >= opts.length) return extraImages.value
+  return null
+})
+
 // 本地答案编辑状态（工作台内存保存，入库时随 question 对象一同传递）
 const editingAnswer = ref(false)
 const editingUserAnswer = ref(false)
@@ -38,13 +74,13 @@ const cancelUserAnswer = () => { editingUserAnswer.value = false }
 
 <template>
   <div
-    class="question-card group relative overflow-hidden rounded-2xl border bg-white/70 p-6 shadow-sm backdrop-blur-xl transition-shadow hover:shadow-md dark:bg-white/[0.03]"
+    class="question-card group relative overflow-hidden rounded-2xl border bg-white/70 p-6 shadow-sm transition-shadow hover:shadow-md dark:bg-white/[0.03]"
     :class="
       selected
         ? 'border-blue-500/50 shadow-blue-500/10 dark:border-indigo-500/50 dark:shadow-indigo-500/20'
         : 'border-slate-200/60 dark:border-white/10 hover:border-blue-200 dark:hover:border-white/15'
     "
-    @click="emit('toggle', question.question_id)"
+    @click="emit('toggle', question.uid)"
   >
     <!-- 选中态背景 -->
     <div
@@ -52,6 +88,15 @@ const cancelUserAnswer = () => { editingUserAnswer.value = false }
       class="absolute inset-0 -z-10 bg-gradient-to-br from-blue-500/[0.03] to-indigo-500/[0.03] dark:from-indigo-500/[0.05] dark:to-purple-500/[0.05]"
     ></div>
 
+    <!-- 大题标签 -->
+    <div
+      v-if="question.section_title"
+      class="mb-4 flex items-center gap-2 border-l-2 border-blue-400/60 pl-3 dark:border-indigo-400/50"
+    >
+      <i class="fa-solid fa-layer-group text-xs text-blue-400 dark:text-indigo-400"></i>
+      <span class="text-xs font-bold tracking-wide text-slate-500 dark:text-slate-400">{{ question.section_title }}</span>
+    </div>
+
     <!-- 顶部状态栏 -->
     <div class="mb-6 flex items-start gap-4">
       <!-- 题型与标签 -->
@@ -97,15 +142,31 @@ const cancelUserAnswer = () => { editingUserAnswer.value = false }
         </div>
       </template>
 
+      <!-- 兜底渲染：无选项配对时独立展示 -->
+      <div v-if="extraImages.length && !optionImages" class="my-4 flex flex-wrap gap-2">
+        <img
+          v-for="(src, i) in extraImages" :key="'extra-' + i"
+          :src="src"
+          class="max-h-[200px] cursor-zoom-in rounded-2xl border border-slate-100 object-contain shadow-sm transition-transform hover:scale-[1.01] dark:border-white/5"
+          @click.stop="() => emit('open-image', src)"
+        />
+      </div>
+
       <!-- 选项 -->
       <div v-if="question.options?.length" class="mt-8 grid gap-3 sm:grid-cols-2">
         <div
           v-for="(opt, idx) in question.options"
           :key="idx"
-          class="flex items-start gap-3 rounded-xl border border-slate-100 bg-slate-50/30 p-4 text-[13px] font-bold text-slate-700 hover:bg-slate-50 dark:border-white/5 dark:bg-white/5 dark:text-slate-300 dark:hover:bg-white/10"
+          class="flex items-center gap-3 rounded-xl border border-slate-100 bg-slate-50/30 p-4 text-[13px] font-bold text-slate-700 hover:bg-slate-50 dark:border-white/5 dark:bg-white/5 dark:text-slate-300 dark:hover:bg-white/10"
         >
           <span class="flex h-5 w-5 shrink-0 items-center justify-center rounded bg-white text-[10px] shadow-sm dark:bg-slate-800">{{ formatOption(opt)[0] }}</span>
           <span class="flex-1">{{ formatOption(opt).slice(2) }}</span>
+          <img
+            v-if="optionImages?.[idx]"
+            :src="optionImages[idx]"
+            class="max-h-[100px] max-w-[160px] shrink-0 cursor-zoom-in rounded-lg border border-slate-100 object-contain dark:border-white/5"
+            @click.stop="() => emit('open-image', optionImages[idx])"
+          />
         </div>
       </div>
     </div>
@@ -124,4 +185,7 @@ const cancelUserAnswer = () => { editingUserAnswer.value = false }
 .question-content th {
   @apply bg-slate-50 font-bold dark:bg-white/[0.04];
 }
+.question-content img {
+  @apply inline-block max-h-[200px] object-contain;
+}
 </style>
diff --git a/frontend/src/components/QuestionDetailModal.vue b/frontend/src/components/question/QuestionDetailModal.vue
similarity index 91%
rename from frontend/src/components/QuestionDetailModal.vue
rename to frontend/src/components/question/QuestionDetailModal.vue
index 7d814594..a5b3113a 100644
--- a/frontend/src/components/QuestionDetailModal.vue
+++ b/frontend/src/components/question/QuestionDetailModal.vue
@@ -1,13 +1,33 @@
 <script setup>
-import { ref, watch, nextTick, onBeforeUnmount } from 'vue'
-import { isHtml, sanitizeHtml, formatOption } from '../utils.js'
-import * as api from '../api.js'
+/**
+ * QuestionDetailModal.vue
+ * 题目详情弹窗（查看题目内容、答案、AI 解题与对话）
+ */
+import { ref, computed, watch, nextTick, onBeforeUnmount } from 'vue'
+import { isHtml, sanitizeHtml, formatOption } from '@/utils.js'
+import * as api from '@/api.js'
 
 const props = defineProps({
   open: { type: Boolean, default: false },
   question: { type: Object, default: null },
 })
 
+const extraImages = computed(() => {
+  const refs = props.question?.image_refs || []
+  if (!refs.length) return []
+  const blocks = props.question?.content_json || []
+  const embedded = new Set(
+    blocks.filter(b => b.block_type === 'image' && b.content).map(b => b.content)
+  )
+  const htmlContent = blocks.filter(b => b.block_type === 'text').map(b => b.content || '').join(' ')
+  return refs.filter(r => {
+    if (embedded.has(r)) return false
+    const filename = r.split('/').pop()
+    if (filename && htmlContent.includes(filename)) return false
+    return true
+  })
+})
+
 const emit = defineEmits(['close', 'open-image', 'deleted', 'answer-saved', 'review-status-changed', 'push-toast', 'start-chat'])
 
 const activeTab = ref('content') // 'content' | 'analysis'
@@ -145,7 +165,7 @@ const reviewStatusOptions = [
     <Transition name="modal-fade">
       <div v-if="open" class="fixed inset-0 z-[100] flex items-center justify-center p-4 sm:p-6">
         <!-- 遮罩层 -->
-        <div class="absolute inset-0 bg-slate-950/60 backdrop-blur-md" @click="emit('close')"></div>
+        <div class="absolute inset-0 bg-slate-950/60" @click="emit('close')"></div>
         
         <!-- 弹窗主体 -->
         <div class="relative flex h-full max-h-[90vh] w-full max-w-5xl flex-col overflow-hidden rounded-[2.5rem] border border-white/10 bg-white shadow-2xl transition-all dark:bg-[#0F111A]">
@@ -197,6 +217,16 @@ const reviewStatusOptions = [
                   <img v-else-if="b.block_type === 'image'" :src="b.content" class="mb-6 max-w-full cursor-zoom-in rounded-2xl border border-slate-200 shadow-sm dark:border-white/10" @click="emit('open-image', b.content)" />
                 </template>
 
+                <!-- 兜底渲染 image_refs 中未嵌入 content_json 的图片 -->
+                <div v-if="extraImages.length" class="mb-6 flex flex-wrap gap-3">
+                  <img
+                    v-for="(src, i) in extraImages" :key="'extra-' + i"
+                    :src="src"
+                    class="max-h-[240px] cursor-zoom-in rounded-2xl border border-slate-200 object-contain shadow-sm dark:border-white/10"
+                    @click="emit('open-image', src)"
+                  />
+                </div>
+
                 <!-- 选项 -->
                 <div v-if="question?.options_json" class="mt-8 grid gap-3">
                   <div v-for="(opt, idx) in question.options_json" :key="idx" class="rounded-2xl border border-slate-100 bg-slate-50/50 p-4 text-sm font-bold text-slate-700 dark:border-white/5 dark:bg-white/5 dark:text-slate-300">
diff --git a/frontend/src/components/QuestionItem.vue b/frontend/src/components/question/QuestionItem.vue
similarity index 70%
rename from frontend/src/components/QuestionItem.vue
rename to frontend/src/components/question/QuestionItem.vue
index 6289ee10..a579ba0c 100644
--- a/frontend/src/components/QuestionItem.vue
+++ b/frontend/src/components/question/QuestionItem.vue
@@ -1,6 +1,10 @@
 <script setup>
+/**
+ * QuestionItem.vue
+ * 题目条目（分割结果列表项）
+ */
 import { ref, watch, nextTick } from 'vue'
-import { getQuestionSnippet, typesetMath } from '../utils.js'
+import { getQuestionSnippet, typesetMath } from '@/utils.js'
 
 const props = defineProps({
   question: { type: Object, required: true },
@@ -50,14 +54,14 @@ const statusIcon = (status) => {
 <template>
   <div
     @click="selectable ? emit('toggle-select', question.id) : emit('click', question)"
-    class="group cursor-pointer rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl transition-all hover:shadow-md dark:border-white/10 dark:bg-white/[0.03]"
-    :class="{ 'ring-2 ring-indigo-500/50 border-indigo-300 dark:border-indigo-500/40': selected }"
+    class="group cursor-pointer rounded-lg border border-white/[0.06] bg-white/[0.02] p-4 transition-all hover:bg-white/[0.04] hover:border-white/[0.1]"
+    :class="{ 'border-[rgb(129,115,223)]/40 bg-[rgb(129,115,223)]/[0.06]': selected }"
   >
     <div class="flex items-start gap-4">
       <!-- 选择复选框 -->
       <div v-if="selectable" class="flex shrink-0 items-center pt-1" @click.stop="emit('toggle-select', question.id)">
         <div class="flex h-5 w-5 items-center justify-center rounded-lg border-2 transition-all"
-          :class="selected ? 'border-indigo-500 bg-indigo-500 text-white' : 'border-slate-300 dark:border-slate-600'">
+          :class="selected ? 'border-[rgb(129,115,223)] bg-[rgb(129,115,223)] text-white' : 'border-white/[0.15]'">
           <i v-if="selected" class="fa-solid fa-check text-xs"></i>
         </div>
       </div>
@@ -67,13 +71,13 @@ const statusIcon = (status) => {
         <div class="mb-2 flex flex-wrap items-center gap-2">
           <!-- 复习状态图标 -->
           <i v-if="showStatus" class="fa-solid text-sm" :class="[statusIcon(question.review_status), statusColor(question.review_status)]"></i>
-          <span class="rounded-full bg-slate-100 px-2 py-1 text-xs font-bold uppercase tracking-widest text-slate-500 dark:bg-white/5 dark:text-slate-400">{{ question.question_type }}</span>
-          <span v-if="question.subject" class="rounded-full bg-blue-50 px-2 py-1 text-xs font-bold text-blue-600 dark:bg-blue-500/10 dark:text-blue-300">{{ question.subject }}</span>
-          <span v-for="tag in tags()" :key="tag" class="rounded-full border border-indigo-500/20 bg-indigo-500/5 px-2 py-1 text-xs font-bold text-indigo-600 dark:text-indigo-300">{{ tag }}</span>
+          <span class="rounded-full bg-white/[0.04] px-2 py-0.5 text-xs font-medium text-[#8a8f98]">{{ question.question_type }}</span>
+          <span v-if="question.subject" class="rounded-full bg-[rgb(129,115,223)]/10 px-2 py-0.5 text-xs font-medium text-[rgb(145,132,235)]">{{ question.subject }}</span>
+          <span v-for="tag in tags()" :key="tag" class="rounded-full border border-white/[0.06] px-2 py-0.5 text-xs font-medium text-[#8a8f98]">{{ tag }}</span>
         </div>
 
         <!-- 摘要 -->
-        <p class="line-clamp-2 text-sm font-bold leading-relaxed text-slate-700 group-hover:text-slate-900 dark:text-slate-300 dark:group-hover:text-white">{{ summary() }}</p>
+        <p class="line-clamp-2 text-sm font-medium leading-relaxed text-[#d0d6e0] group-hover:text-[#f7f8f8]">{{ summary() }}</p>
 
         <!-- 题目图片 -->
         <div v-if="question.content_json?.some(b => b.block_type === 'image' && b.content)" class="mt-3 flex flex-wrap gap-2">
@@ -81,7 +85,7 @@ const statusIcon = (status) => {
             v-for="(b, i) in question.content_json.filter(b => b.block_type === 'image' && b.content)"
             :key="i"
             :src="b.content"
-            class="max-h-32 rounded-xl border border-slate-100 object-contain dark:border-white/5"
+            class="max-h-32 rounded-lg border border-white/[0.06] object-contain"
             @click.stop
           />
         </div>
@@ -91,9 +95,9 @@ const statusIcon = (status) => {
           <div
             v-for="(opt, idx) in question.options_json"
             :key="idx"
-            class="flex items-start gap-2 rounded-lg border border-slate-100 bg-slate-50/50 px-3 py-1.5 text-xs font-bold text-slate-600 dark:border-white/5 dark:bg-white/[0.02] dark:text-slate-400"
+            class="flex items-start gap-2 rounded-md border border-white/[0.06] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#8a8f98]"
           >
-            <span class="shrink-0 text-slate-400 dark:text-slate-500">{{ String.fromCharCode(65 + idx) }}.</span>
+            <span class="shrink-0 text-[#62666d]">{{ String.fromCharCode(65 + idx) }}.</span>
             <span class="line-clamp-1">{{ String(opt).replace(/^[A-Da-d][.、．]\s*/, '') }}</span>
           </div>
         </div>
diff --git a/frontend/src/components/QuestionItemSkeleton.vue b/frontend/src/components/question/QuestionItemSkeleton.vue
similarity index 91%
rename from frontend/src/components/QuestionItemSkeleton.vue
rename to frontend/src/components/question/QuestionItemSkeleton.vue
index 9101ca3c..0e1c914c 100644
--- a/frontend/src/components/QuestionItemSkeleton.vue
+++ b/frontend/src/components/question/QuestionItemSkeleton.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * QuestionItemSkeleton.vue
+ * 题目条目骨架屏
+ */
 defineProps({
   count: { type: Number, default: 5 },
 })
@@ -8,7 +12,7 @@ defineProps({
   <div class="space-y-4">
     <div
       v-for="i in count" :key="i"
-      class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]"
+      class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm dark:border-white/10 dark:bg-white/[0.03]"
     >
       <div class="flex items-start gap-4">
         <!-- 内容区：精确复刻 QuestionItem 的 min-w-0 flex-1 -->
diff --git a/frontend/src/components/QuestionList.vue b/frontend/src/components/question/QuestionList.vue
similarity index 91%
rename from frontend/src/components/QuestionList.vue
rename to frontend/src/components/question/QuestionList.vue
index bc0f046d..de3a17f6 100644
--- a/frontend/src/components/QuestionList.vue
+++ b/frontend/src/components/question/QuestionList.vue
@@ -1,7 +1,11 @@
 <script setup>
+/**
+ * QuestionList.vue
+ * 题目列表容器（选择 + 批量操作）
+ */
 import { ref, nextTick, watch, onMounted } from 'vue'
 import QuestionCard from './QuestionCard.vue'
-import { typesetMath } from '../utils.js'
+import { typesetMath } from '@/utils.js'
 
 const props = defineProps({
   questions: { type: Array, default: () => [] },
@@ -64,9 +68,9 @@ defineExpose({ initSortable, questionsBoxEl, triggerTypeset })
       <div ref="questionListEl" class="relative z-10 mt-6 grid gap-5" id="questionList">
         <QuestionCard
           v-for="q in questions"
-          :key="q.question_id"
+          :key="q.uid"
           :question="q"
-          :selected="selectedIds.has(q.question_id)"
+          :selected="selectedIds.has(q.uid)"
           @toggle="(id) => emit('toggle-select', id)"
           @open-image="(src) => emit('open-image', src)"
         />
diff --git a/frontend/src/components/AiAnalysisModal.vue b/frontend/src/components/review/AiAnalysisModal.vue
similarity index 98%
rename from frontend/src/components/AiAnalysisModal.vue
rename to frontend/src/components/review/AiAnalysisModal.vue
index 638f1040..840796f3 100644
--- a/frontend/src/components/AiAnalysisModal.vue
+++ b/frontend/src/components/review/AiAnalysisModal.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * AiAnalysisModal.vue
+ * AI 分析弹窗（知识点讲解）
+ */
 const props = defineProps({
   open: { type: Boolean, default: false },
   loading: { type: Boolean, default: false },
@@ -13,7 +17,7 @@ const emit = defineEmits(['close'])
     <Transition name="modal-fade">
       <div v-if="open" class="fixed inset-0 z-[100] flex items-center justify-center p-4 sm:p-6">
         <!-- 遮罩 -->
-        <div class="absolute inset-0 bg-slate-950/60 backdrop-blur-md" @click="emit('close')"></div>
+        <div class="absolute inset-0 bg-slate-950/60" @click="emit('close')"></div>
 
         <!-- 弹窗主体 -->
         <div class="relative flex max-h-[85vh] w-full max-w-3xl flex-col overflow-hidden rounded-[2rem] border border-white/10 bg-white shadow-2xl dark:bg-[#0F111A]">
diff --git a/frontend/src/components/ProviderDialog.vue b/frontend/src/components/settings/ProviderDialog.vue
similarity index 95%
rename from frontend/src/components/ProviderDialog.vue
rename to frontend/src/components/settings/ProviderDialog.vue
index b8b92adc..11aa32d9 100644
--- a/frontend/src/components/ProviderDialog.vue
+++ b/frontend/src/components/settings/ProviderDialog.vue
@@ -1,7 +1,12 @@
 <script setup>
-import { ref, computed, watch, inject } from 'vue'
+/**
+ * ProviderDialog.vue
+ * API Provider 配置弹窗
+ */
+import { ref, computed, watch } from 'vue'
+import { useToast } from '@/composables/useToast.js'
 
-const pushToast = inject('pushToast', (type, msg) => { console.warn(`[${type}] ${msg}`) })
+const { pushToast } = useToast()
 
 const props = defineProps({
   open: { type: Boolean, default: false },
@@ -202,8 +207,8 @@ const selectOption = (field, value) => {
   <Teleport to="body">
     <!-- 背景遮罩 -->
     <div
-      class="fixed inset-0 z-[100] transition-[background-color,backdrop-filter] duration-200 md:left-64"
-      :class="open ? 'bg-black/40 backdrop-blur-sm pointer-events-auto' : 'bg-transparent backdrop-blur-none pointer-events-none'"
+      class="fixed inset-0 z-[100] transition-colors duration-200 md:left-64"
+      :class="open ? 'bg-black/40 pointer-events-auto' : 'bg-transparent pointer-events-none'"
       @click="emit('close')"
     ></div>
 
@@ -284,7 +289,7 @@ const selectOption = (field, value) => {
                   {{ type === 'paddleocr' ? 'OCR 模型' : '默认模型' }}
                   <span v-if="type !== 'paddleocr'" class="group relative">
                     <i class="fa-solid fa-circle-info cursor-help text-slate-400 transition-colors hover:text-blue-500 dark:text-slate-500"></i>
-                    <span class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded-lg border border-slate-200/60 bg-white/90 px-3 py-1.5 text-xs font-normal text-slate-600 opacity-0 shadow-lg backdrop-blur-xl transition-opacity group-hover:opacity-100 dark:border-white/10 dark:bg-[#0A0A0F]/90 dark:text-slate-300">
+                    <span class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded-lg border border-slate-200/60 bg-white/90 px-3 py-1.5 text-xs font-normal text-slate-600 opacity-0 shadow-lg transition-opacity group-hover:opacity-100 dark:border-white/10 dark:bg-[#0A0A0F]/90 dark:text-slate-300">
                       用于题目分割、纠错等核心任务
                     </span>
                   </span>
@@ -294,14 +299,14 @@ const selectOption = (field, value) => {
                   <button
                     type="button"
                     @click.stop="toggleDropdown('model_name')"
-                    class="flex w-full items-center justify-between rounded-xl border border-slate-200/80 bg-white/70 px-4 py-2.5 text-left text-sm backdrop-blur-xl transition-colors dark:border-white/10 dark:bg-slate-800/60"
+                    class="flex w-full items-center justify-between rounded-xl border border-slate-200/80 bg-white/70 px-4 py-2.5 text-left text-sm transition-colors dark:border-white/10 dark:bg-slate-800/60"
                     :class="form.model_name ? 'text-slate-800 dark:text-slate-200' : 'text-slate-400 dark:text-slate-500'"
                   >
                     <span class="truncate">{{ form.model_name || '请选择模型' }}</span>
                     <i class="fa-solid fa-chevron-down ml-2 text-[10px] text-slate-400 transition-transform" :class="openDropdown === 'model_name' ? 'rotate-180' : ''"></i>
                   </button>
                   <Transition name="dropdown">
-                    <div v-if="openDropdown === 'model_name'" class="absolute z-50 mt-1.5 max-h-48 w-full overflow-y-auto rounded-xl border border-slate-200/60 bg-white/80 py-1 shadow-xl backdrop-blur-xl dark:border-white/10 dark:bg-[#0A0A0F]/80">
+                    <div v-if="openDropdown === 'model_name'" class="absolute z-50 mt-1.5 max-h-48 w-full overflow-y-auto rounded-xl border border-slate-200/60 bg-white/80 py-1 shadow-xl dark:border-white/10 dark:bg-[#0A0A0F]/80">
                       <button
                         v-for="m in modelList" :key="m"
                         type="button"
@@ -330,7 +335,7 @@ const selectOption = (field, value) => {
                   <span class="font-normal text-slate-400">（可选）</span>
                   <span class="group relative">
                     <i class="fa-solid fa-circle-info cursor-help text-slate-400 transition-colors hover:text-blue-500 dark:text-slate-500"></i>
-                    <span class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded-lg border border-slate-200/60 bg-white/90 px-3 py-1.5 text-xs font-normal text-slate-600 opacity-0 shadow-lg backdrop-blur-xl transition-opacity group-hover:opacity-100 dark:border-white/10 dark:bg-[#0A0A0F]/90 dark:text-slate-300">
+                    <span class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded-lg border border-slate-200/60 bg-white/90 px-3 py-1.5 text-xs font-normal text-slate-600 opacity-0 shadow-lg transition-opacity group-hover:opacity-100 dark:border-white/10 dark:bg-[#0A0A0F]/90 dark:text-slate-300">
                       用于科目识别等简单任务，更快更省 Token
                     </span>
                   </span>
@@ -339,14 +344,14 @@ const selectOption = (field, value) => {
                   <button
                     type="button"
                     @click.stop="toggleDropdown('light_model_name')"
-                    class="flex w-full items-center justify-between rounded-xl border border-slate-200/80 bg-white/70 px-4 py-2.5 text-left text-sm backdrop-blur-xl transition-colors dark:border-white/10 dark:bg-slate-800/60"
+                    class="flex w-full items-center justify-between rounded-xl border border-slate-200/80 bg-white/70 px-4 py-2.5 text-left text-sm transition-colors dark:border-white/10 dark:bg-slate-800/60"
                     :class="form.light_model_name ? 'text-slate-800 dark:text-slate-200' : 'text-slate-400 dark:text-slate-500'"
                   >
                     <span class="truncate">{{ form.light_model_name || '不使用' }}</span>
                     <i class="fa-solid fa-chevron-down ml-2 text-[10px] text-slate-400 transition-transform" :class="openDropdown === 'light_model_name' ? 'rotate-180' : ''"></i>
                   </button>
                   <Transition name="dropdown">
-                    <div v-if="openDropdown === 'light_model_name'" class="absolute z-50 mt-1.5 max-h-48 w-full overflow-y-auto rounded-xl border border-slate-200/60 bg-white/80 py-1 shadow-xl backdrop-blur-xl dark:border-white/10 dark:bg-[#0A0A0F]/80">
+                    <div v-if="openDropdown === 'light_model_name'" class="absolute z-50 mt-1.5 max-h-48 w-full overflow-y-auto rounded-xl border border-slate-200/60 bg-white/80 py-1 shadow-xl dark:border-white/10 dark:bg-[#0A0A0F]/80">
                       <button
                         type="button"
                         @click.stop="selectOption('light_model_name', '')"
diff --git a/frontend/src/components/ProviderSection.vue b/frontend/src/components/settings/ProviderSection.vue
similarity index 94%
rename from frontend/src/components/ProviderSection.vue
rename to frontend/src/components/settings/ProviderSection.vue
index b399c787..6f9f3afd 100644
--- a/frontend/src/components/ProviderSection.vue
+++ b/frontend/src/components/settings/ProviderSection.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * ProviderSection.vue
+ * API Provider 配置区块
+ */
 defineProps({
   icon: { type: String, required: true },
   title: { type: String, required: true },
@@ -42,7 +46,7 @@ const emit = defineEmits(['add', 'toggle-active', 'edit', 'remove'])
     <div class="space-y-2">
       <div
         v-for="(provider, idx) in providers" :key="provider.id"
-        class="flex cursor-pointer items-center gap-3 rounded-2xl border border-slate-200/60 bg-white/70 px-5 py-3.5 shadow-sm backdrop-blur-xl transition-all hover:border-blue-300 dark:border-white/10 dark:bg-white/[0.03] dark:hover:border-blue-500/30"
+        class="flex cursor-pointer items-center gap-3 rounded-2xl border border-slate-200/60 bg-white/70 px-5 py-3.5 shadow-sm transition-all hover:border-blue-300 dark:border-white/10 dark:bg-white/[0.03] dark:hover:border-blue-500/30"
         @click="emit('toggle-active', provider.id)"
       >
         <!-- 激活单选 -->
diff --git a/frontend/src/components/ActionBar.vue b/frontend/src/components/workspace/ActionBar.vue
similarity index 54%
rename from frontend/src/components/ActionBar.vue
rename to frontend/src/components/workspace/ActionBar.vue
index 7b7a327b..e75e394c 100644
--- a/frontend/src/components/ActionBar.vue
+++ b/frontend/src/components/workspace/ActionBar.vue
@@ -1,10 +1,16 @@
 <script setup>
+/**
+ * ActionBar.vue
+ * 操作栏（导出、保存、取消等按钮）
+ */
 
-defineProps({
+const props = defineProps({
   splitEnabled: { type: Boolean, default: false },
   exportEnabled: { type: Boolean, default: false },
   splitting: { type: Boolean, default: false },
   splitCompleted: { type: Boolean, default: false },
+  uploadMode: { type: String, default: 'exam' },
+  eraseEnabled: { type: Boolean, default: false },
 })
 
 const emit = defineEmits(['split', 'export', 'save-to-db'])
@@ -14,42 +20,36 @@ const emit = defineEmits(['split', 'export', 'save-to-db'])
   <div class="flex items-center justify-center gap-4">
     <!-- 主操作区 -->
     <div class="flex items-center gap-3">
-    <!-- 主操作按钮：引入超级炫酷的炫彩发光效果 -->
+    <!-- 主操作按钮 -->
     <button
       type="button"
       class="group relative inline-flex h-14 items-center justify-center sm:w-auto w-full disabled:cursor-not-allowed disabled:opacity-50"
       :disabled="!splitEnabled"
       @click="emit('split')"
     >
-      <!-- 背景光晕 (悬浮时放大并提亮) -->
-      <div
-        class="absolute -inset-1 rounded-2xl bg-gradient-to-r from-blue-600 via-indigo-500 to-purple-600 blur-md transition-all duration-500 opacity-0 group-hover:opacity-60 group-hover:blur-xl"
-        :class="!splitEnabled && 'hidden'"
-      ></div>
-      
       <!-- 按钮本体 -->
-      <span class="relative inline-flex h-full w-full items-center justify-center gap-3 rounded-2xl bg-slate-900 px-10 text-[13px] font-black tracking-widest text-white shadow-2xl group-hover:-translate-y-0.5 group-active:translate-y-0 dark:bg-white dark:text-slate-900">
+      <span class="relative overflow-hidden inline-flex h-full w-full items-center justify-center gap-2 rounded-md px-8 text-sm font-medium text-white brand-btn">
         <template v-if="splitting">
-          <i class="fa-solid fa-spinner fa-spin text-blue-400"></i>
-          <span>正在智能分割...</span>
+          <i class="fa-solid fa-spinner fa-spin" :class="uploadMode === 'note' ? 'text-emerald-400' : 'text-white/80'"></i>
+          <span>{{ uploadMode === 'note' ? '正在整理笔记...' : '正在智能分割...' }}</span>
         </template>
         <template v-else-if="splitCompleted">
           <i class="fa-solid fa-check-double text-emerald-400"></i>
-          <span>解析已完成</span>
+          <span>{{ uploadMode === 'note' ? '整理已完成' : '解析已完成' }}</span>
         </template>
         <template v-else>
-          <i class="fa-solid fa-wand-magic-sparkles text-blue-400 group-hover:animate-pulse"></i>
-          <span>启动 AI 智能分割</span>
+          <i class="fa-solid text-white/80" :class="uploadMode === 'note' ? 'fa-book-open' : eraseEnabled ? 'fa-eraser' : 'fa-eye'"></i>
+          <span>{{ uploadMode === 'note' ? '启动 AI 笔记整理' : eraseEnabled ? '开始擦除笔迹' : '开始 OCR 识别' }}</span>
         </template>
       </span>
     </button>
 
-    <!-- 导出按钮：极简现代感 -->
+    <!-- 导出按钮 -->
     <Transition name="action-btn">
       <button
         v-if="exportEnabled"
         type="button"
-        class="group relative inline-flex h-14 items-center justify-center gap-3 rounded-2xl border border-slate-200 bg-white px-8 text-[13px] font-black tracking-widest text-slate-900 shadow-sm hover:border-slate-900 hover:bg-slate-50 disabled:cursor-not-allowed disabled:opacity-30 dark:border-white/10 dark:bg-slate-900 dark:text-white dark:hover:border-white dark:hover:bg-slate-800"
+        class="group relative inline-flex h-14 items-center justify-center gap-3 rounded-lg brand-btn px-8 text-sm font-medium tracking-widest text-[#f7f8f8] hover:border-white/[0.12] hover:bg-white/[0.05] disabled:cursor-not-allowed disabled:opacity-30"
         @click="emit('export')"
       >
         <i class="fa-solid fa-file-export transition-transform group-hover:-translate-x-0.5"></i>
@@ -57,12 +57,12 @@ const emit = defineEmits(['split', 'export', 'save-to-db'])
       </button>
     </Transition>
 
-    <!-- 导入错题库按钮：极简现代感 -->
+    <!-- 导入错题库按钮 -->
     <Transition name="action-btn">
       <button
         v-if="exportEnabled"
         type="button"
-        class="group relative inline-flex h-14 items-center justify-center gap-3 rounded-2xl border border-slate-200 bg-white px-8 text-[13px] font-black tracking-widest text-slate-900 shadow-sm hover:border-slate-900 hover:bg-slate-50 disabled:cursor-not-allowed disabled:opacity-30 dark:border-white/10 dark:bg-slate-900 dark:text-white dark:hover:border-white dark:hover:bg-slate-800"
+        class="group relative inline-flex h-14 items-center justify-center gap-3 rounded-lg brand-btn px-8 text-sm font-medium tracking-widest text-[#f7f8f8] hover:border-white/[0.12] hover:bg-white/[0.05] disabled:cursor-not-allowed disabled:opacity-30"
         @click="emit('save-to-db')"
       >
         <i class="fa-solid fa-database transition-transform group-hover:-translate-y-0.5"></i>
@@ -89,4 +89,4 @@ const emit = defineEmits(['split', 'export', 'save-to-db'])
   opacity: 0;
   transform: translateY(8px) scale(0.95);
 }
-</style>
\ No newline at end of file
+</style>
diff --git a/frontend/src/components/workspace/AnswerInputModal.vue b/frontend/src/components/workspace/AnswerInputModal.vue
new file mode 100644
index 00000000..b35756e9
--- /dev/null
+++ b/frontend/src/components/workspace/AnswerInputModal.vue
@@ -0,0 +1,48 @@
+<script setup>
+/**
+ * AnswerInputModal.vue
+ * AI 辅导前置 — 答案录入弹窗
+ */
+import { defineProps, defineEmits } from 'vue'
+
+const props = defineProps({
+  open: Boolean,
+  text: String,
+  saving: Boolean,
+})
+const emit = defineEmits(['update:open', 'update:text', 'confirm'])
+</script>
+
+<template>
+  <div v-if="open" class="fixed inset-0 z-[100] flex items-center justify-center p-4">
+    <div class="absolute inset-0 bg-black/60" @click="emit('update:open', false)"></div>
+    <div class="relative w-full max-w-lg rounded-lg brand-btn p-6">
+      <h3 class="mb-1 text-base font-medium text-[#f7f8f8]">录入答案</h3>
+      <p class="mb-4 text-xs text-[#62666d]">
+        AI 辅导需要正确答案作为参考。支持 Markdown 格式，数学公式使用 LaTeX（$..$ 行内，$$...$$ 独占行）
+      </p>
+      <textarea
+        :value="text"
+        @input="emit('update:text', $event.target.value)"
+        rows="10"
+        placeholder="在此粘贴或输入答案/解析..."
+        class="w-full resize-none rounded-md border border-white/[0.08] bg-white/[0.02] px-4 py-3 font-mono text-sm text-[#d0d6e0] placeholder-[#62666d] focus:border-white/[0.12] focus:outline-none transition-colors"
+      ></textarea>
+      <div class="mt-4 flex justify-end gap-3">
+        <button
+          @click="emit('update:open', false)"
+          class="rounded-md border border-white/[0.08] bg-white/[0.02] px-4 py-2 text-sm font-medium text-[#d0d6e0] transition-colors hover:bg-white/[0.05]"
+        >
+          取消
+        </button>
+        <button
+          @click="emit('confirm')"
+          :disabled="saving"
+          class="rounded-md bg-[#5e6ad2] px-4 py-2 text-sm font-medium text-white transition-colors hover:bg-[#7170ff] disabled:opacity-50"
+        >
+          {{ saving ? '保存中...' : '保存并开始辅导' }}
+        </button>
+      </div>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/ContentPanel.vue b/frontend/src/components/workspace/ContentPanel.vue
new file mode 100644
index 00000000..8ab557f1
--- /dev/null
+++ b/frontend/src/components/workspace/ContentPanel.vue
@@ -0,0 +1,77 @@
+<script setup>
+/**
+ * ContentPanel.vue
+ * 右侧内容面板布局 — Linear 风格
+ * 支持：顶部栏（标题 + 步骤 tab + 工具栏）+ 主内容区 + 可选右侧属性栏
+ */
+defineProps({
+  title: { type: String, required: true },
+  steps: { type: Array, default: () => [] },  // [{ label, active, done }]
+  currentStep: { type: Number, default: -1 }, // -1 = 不显示步骤
+})
+
+const emit = defineEmits(['step-click'])
+</script>
+
+<template>
+  <div class="h-full flex flex-col rounded-none md:rounded-lg brand-btn overflow-hidden" style="background: rgba(255,255,255,0.04);">
+    <!-- 顶部栏 -->
+    <header class="flex items-center h-14 px-4 border-b border-white/[0.08] shrink-0 gap-4">
+      <!-- 标题 -->
+      <h2 class="text-sm font-medium text-[#f7f8f8] shrink-0">{{ title }}</h2>
+
+      <!-- 步骤 tab（可选） -->
+      <div v-if="steps.length" class="flex items-center gap-1 ml-2">
+        <button
+          v-for="(s, i) in steps"
+          :key="i"
+          @click="emit('step-click', i)"
+          class="flex items-center gap-1.5 px-2.5 py-1 rounded-md text-xs transition-colors"
+          :class="i === currentStep
+            ? 'bg-white/[0.06] text-[#f7f8f8] font-medium'
+            : s.done
+              ? 'text-[#8a8f98] hover:bg-white/[0.04] hover:text-[#d0d6e0]'
+              : 'text-[#62666d] cursor-default'"
+        >
+          <span
+            class="flex h-4 w-4 items-center justify-center rounded text-[10px]"
+            :class="s.done
+              ? 'bg-[rgb(129,115,223)] text-white'
+              : i === currentStep
+                ? 'bg-white/[0.08] text-[#f7f8f8]'
+                : 'bg-white/[0.04] text-[#62666d]'"
+          >
+            <i v-if="s.done" class="fa-solid fa-check text-[8px]"></i>
+            <span v-else>{{ i + 1 }}</span>
+          </span>
+          <span>{{ s.label }}</span>
+        </button>
+      </div>
+
+      <!-- 左侧附加操作 -->
+      <div v-if="$slots.actions" class="flex items-center gap-1 text-[#62666d]">
+        <slot name="actions"></slot>
+      </div>
+
+      <!-- 右侧工具栏 -->
+      <div class="flex shrink-0 items-center gap-3 ml-auto">
+        <!-- 子组件可通过 Teleport 注入内容 -->
+        <div :id="'panel-toolbar-' + title" class="flex items-center gap-3"></div>
+        <slot name="toolbar"></slot>
+      </div>
+    </header>
+
+    <!-- 内容区：主体 + 可选右侧栏 -->
+    <div class="flex-1 flex overflow-hidden">
+      <!-- 主内容 -->
+      <div class="flex-1 min-h-0 overflow-y-auto px-4 py-3 flex flex-col">
+        <slot></slot>
+      </div>
+
+      <!-- 右侧属性栏（可选） -->
+      <aside v-if="$slots.sidebar" class="hidden lg:flex w-80 shrink-0 flex-col border-l border-white/[0.05] overflow-y-auto">
+        <slot name="sidebar"></slot>
+      </aside>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/ErasePreview.vue b/frontend/src/components/workspace/ErasePreview.vue
new file mode 100644
index 00000000..bec1f4bc
--- /dev/null
+++ b/frontend/src/components/workspace/ErasePreview.vue
@@ -0,0 +1,98 @@
+<script setup>
+/**
+ * ErasePreview.vue
+ * 擦除前后对比预览 — 拖拽滑块对比，单张显示 + 分页
+ */
+import { ref, computed } from 'vue'
+import SplitLoading from './SplitLoading.vue'
+
+const props = defineProps({
+  images: { type: Array, default: () => [] },
+  loading: { type: Boolean, default: false },
+  previewUrl: { type: String, default: '' },
+})
+
+// 当前页码（0-based）
+const currentPage = ref(0)
+const currentImg = computed(() => props.images[currentPage.value])
+
+// 滑块位置（0-100）
+const sliderPos = ref(50)
+
+function onPointerDown(event) {
+  const container = event.currentTarget.closest('.compare-container')
+  if (!container) return
+
+  const onMove = (e) => {
+    const rect = container.getBoundingClientRect()
+    const x = (e.clientX ?? e.touches?.[0]?.clientX ?? 0) - rect.left
+    sliderPos.value = Math.max(0, Math.min(100, (x / rect.width) * 100))
+  }
+  const onUp = () => {
+    document.removeEventListener('pointermove', onMove)
+    document.removeEventListener('pointerup', onUp)
+  }
+  document.addEventListener('pointermove', onMove)
+  document.addEventListener('pointerup', onUp)
+}
+</script>
+
+<template>
+  <div class="flex flex-col h-full">
+    <!-- 加载中 -->
+    <div v-if="loading" class="flex-1 min-h-0 relative">
+      <img v-if="previewUrl" :src="previewUrl" class="absolute inset-0 w-full h-full object-contain opacity-10 blur-sm" alt="" />
+      <SplitLoading title="正在擦除手写笔迹" subtitle="EnsExam 模型正在识别并移除手写内容" />
+    </div>
+
+    <!-- 无数据 -->
+    <div v-else-if="!images.length" class="flex-1 flex items-center justify-center">
+      <p class="text-sm text-[#62666d]">暂无擦除数据</p>
+    </div>
+
+    <!-- 预览（单张 + 分页） -->
+    <template v-else>
+      <div
+        class="compare-container flex-1 min-h-0 relative select-none overflow-hidden"
+        style="touch-action: none;"
+      >
+        <!-- 擦除后（底层） -->
+        <img :src="currentImg.erased_url" class="block w-full h-full object-contain" alt="" draggable="false" />
+
+        <!-- 原图（上层，clip 裁剪） -->
+        <div
+          class="absolute inset-0 overflow-hidden"
+          :style="{ clipPath: `inset(0 ${100 - sliderPos}% 0 0)` }"
+        >
+          <img :src="currentImg.original_url" class="block w-full h-full object-contain" alt="" draggable="false" />
+        </div>
+
+        <!-- 滑块线 -->
+        <div
+          class="absolute top-0 bottom-0 w-0.5 z-10"
+          style="background: rgb(129,115,223);"
+          :style="{ left: sliderPos + '%' }"
+        >
+          <div
+            class="absolute top-1/2 left-1/2 -translate-x-1/2 -translate-y-1/2 flex h-8 w-8 cursor-ew-resize items-center justify-center rounded-full shadow-lg"
+            style="background: rgb(129,115,223);"
+            @pointerdown.prevent="onPointerDown"
+          >
+            <i class="fa-solid fa-arrows-left-right text-xs text-white"></i>
+          </div>
+        </div>
+      </div>
+
+      <!-- 分页（多张时显示） -->
+      <div v-if="images.length > 1" class="flex items-center justify-center gap-2 py-2 shrink-0">
+        <button
+          v-for="(_, i) in images"
+          :key="i"
+          @click="currentPage = i; sliderPos = 50"
+          class="h-2 rounded-full transition-all"
+          :class="i === currentPage ? 'w-6 bg-[rgb(129,115,223)]' : 'w-2 bg-white/20 hover:bg-white/40'"
+        />
+      </div>
+    </template>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/ErrorBankFilterPanel.vue b/frontend/src/components/workspace/ErrorBankFilterPanel.vue
new file mode 100644
index 00000000..4685abae
--- /dev/null
+++ b/frontend/src/components/workspace/ErrorBankFilterPanel.vue
@@ -0,0 +1,56 @@
+<script setup>
+/**
+ * ErrorBankFilterPanel.vue
+ * 错题库筛选侧边栏
+ */
+import BaseSelect from '@/components/base/BaseSelect.vue'
+
+const props = defineProps({
+  filters: Object,
+  subjects: Array,
+  questionTypes: Array,
+  tagNames: Array,
+  selectedTags: Set,
+})
+
+const emit = defineEmits(['close', 'toggle-tag'])
+</script>
+
+<template>
+  <div class="p-4 space-y-4">
+    <div class="flex items-center justify-between">
+      <span class="text-xs font-medium text-[#f7f8f8]">筛选设置</span>
+      <button @click="emit('close')" class="text-[#62666d] hover:text-[#8a8f98] transition-colors">
+        <i class="fa-solid fa-xmark text-xs"></i>
+      </button>
+    </div>
+
+    <div>
+      <label class="mb-1.5 block text-xs font-medium text-[#62666d]">学科</label>
+      <BaseSelect v-model="filters.subject" :options="subjects" placeholder="全部学科" />
+    </div>
+
+    <div>
+      <label class="mb-1.5 block text-xs font-medium text-[#62666d]">题型</label>
+      <BaseSelect v-model="filters.question_type" :options="questionTypes" placeholder="全部题型" />
+    </div>
+
+    <div>
+      <label class="mb-1.5 block text-xs font-medium text-[#62666d]">复习状态</label>
+      <BaseSelect v-model="filters.review_status" :options="['待复习', '复习中', '已掌握']" placeholder="全部状态" />
+    </div>
+
+    <div v-if="tagNames?.length">
+      <label class="mb-1.5 block text-xs font-medium text-[#62666d]">知识点</label>
+      <div class="flex flex-wrap gap-1.5">
+        <button v-for="tag in tagNames" :key="tag"
+          @click="emit('toggle-tag', tag)"
+          class="rounded-md px-2 py-0.5 text-xs font-medium transition-all"
+          :class="selectedTags?.has(tag)
+            ? 'bg-[rgb(129,115,223)] text-white'
+            : 'border border-white/[0.06] bg-white/[0.02] text-[#62666d] hover:text-[#8a8f98] hover:border-white/[0.1]'"
+        >{{ tag }}</button>
+      </div>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/FileList.vue b/frontend/src/components/workspace/FileList.vue
new file mode 100644
index 00000000..73713168
--- /dev/null
+++ b/frontend/src/components/workspace/FileList.vue
@@ -0,0 +1,69 @@
+<script setup>
+/**
+ * FileList.vue
+ * 已上传文件列表
+ */
+defineProps({
+  pendingFiles: { type: Array, default: () => [] },
+  fileProgress: { type: Object, default: () => ({}) },
+  waitingKeys: { type: Object, default: () => new Set() },
+  uploadBusy: { type: Boolean, default: false },
+  uploadReady: { type: Boolean, default: false },
+  splitting: { type: Boolean, default: false },
+  splitCompleted: { type: Boolean, default: false },
+})
+const emit = defineEmits(['remove-file'])
+</script>
+
+<template>
+  <Transition
+    enter-active-class="transition-all duration-500 ease-out"
+    enter-from-class="opacity-0 -translate-y-4 max-h-0"
+    enter-to-class="opacity-100 translate-y-0 max-h-40"
+    leave-active-class="transition-all duration-300 ease-in"
+    leave-from-class="opacity-100 translate-y-0 max-h-40"
+    leave-to-class="opacity-0 -translate-y-4 max-h-0"
+  >
+    <div v-if="pendingFiles.length" class="relative z-20 w-full overflow-hidden shrink-0">
+      <div class="flex flex-wrap items-center gap-2">
+        <div
+          v-for="item in pendingFiles"
+          :key="item.key"
+          class="relative flex items-center gap-2.5 overflow-hidden rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-2"
+        >
+          <!-- 进度条背景 -->
+          <div
+            class="absolute inset-0 -z-10 bg-[rgba(129,115,223,0.12)] transition-[width] duration-500 ease-out"
+            :style="{ width: `${fileProgress[item.key] ?? 0}%` }"
+          ></div>
+
+          <i
+            class="fa-solid text-sm shrink-0"
+            :class="
+              item.file.name.toLowerCase().endsWith('.pdf')
+                ? 'fa-file-pdf text-rose-400'
+                : 'fa-file-image text-[rgb(129,115,223)]'
+            "
+          ></i>
+          <div class="flex flex-col min-w-0 pr-4 text-left">
+            <span class="truncate text-xs font-medium text-[#f7f8f8]" :title="item.file.name">
+              {{ item.file.name }}
+            </span>
+            <span class="text-[10px] text-[#62666d]">
+              {{ waitingKeys.has(item.key) && uploadBusy ? '等待中' : `${Math.round(fileProgress[item.key] ?? 0)}%` }}
+            </span>
+          </div>
+
+          <button
+            type="button"
+            class="absolute right-1.5 top-1.5 flex h-4 w-4 items-center justify-center rounded text-[#62666d] transition-colors hover:text-rose-400"
+            :disabled="splitting || splitCompleted"
+            @click.stop="() => emit('remove-file', item.key)"
+          >
+            <i class="fa-solid fa-xmark text-[8px]"></i>
+          </button>
+        </div>
+      </div>
+    </div>
+  </Transition>
+</template>
diff --git a/frontend/src/components/workspace/FileUploader.vue b/frontend/src/components/workspace/FileUploader.vue
new file mode 100644
index 00000000..cd0fc30b
--- /dev/null
+++ b/frontend/src/components/workspace/FileUploader.vue
@@ -0,0 +1,89 @@
+<script setup>
+/**
+ * FileUploader.vue
+ * 文件上传拖拽区
+ */
+import { ref } from 'vue'
+
+const props = defineProps({
+  pendingFiles: { type: Array, default: () => [] },
+  fileProgress: { type: Object, default: () => ({}) },
+  waitingKeys: { type: Object, default: () => new Set() },
+  uploadBusy: { type: Boolean, default: false },
+  uploadReady: { type: Boolean, default: false },
+  splitting: { type: Boolean, default: false },
+  splitCompleted: { type: Boolean, default: false },
+  expand: { type: Boolean, default: false },
+  disabled: { type: Boolean, default: false },
+})
+
+const emit = defineEmits(['upload', 'remove-file'])
+
+const fileInputEl = ref(null)
+const uploadHover = ref(false)
+
+const onFileInput = (e) => {
+  if (props.disabled) return
+  emit('upload', e.target.files)
+  e.target.value = ''
+}
+
+const onDrop = (e) => {
+  e.preventDefault()
+  uploadHover.value = false
+  if (props.disabled) return
+  emit('upload', e.dataTransfer.files)
+}
+
+const onClickZone = () => {
+  if (props.disabled) return
+  fileInputEl.value?.click()
+}
+</script>
+
+<template>
+  <div :class="{ 'flex flex-col flex-1': expand }">
+    <div
+      class="group relative flex flex-col items-center justify-center rounded-md border transition-colors"
+      :class="[
+        uploadHover
+          ? 'border-[rgb(129,115,223)]/40 bg-white/[0.04]'
+          : 'border-dashed border-white/[0.08] hover:border-white/[0.12] hover:bg-white/[0.02]',
+        expand ? 'py-14' : 'py-8',
+        disabled ? 'opacity-40 cursor-not-allowed' : 'cursor-pointer'
+      ]"
+      @dragenter.prevent="uploadHover = true"
+      @dragover.prevent="uploadHover = true"
+      @dragleave.prevent="uploadHover = false"
+      @drop="onDrop"
+      @click="onClickZone"
+      @keydown.enter.prevent="onClickZone"
+      @keydown.space.prevent="onClickZone"
+      role="button"
+      tabindex="0"
+    >
+      <div class="flex flex-col items-center gap-3">
+        <i class="fa-solid fa-cloud-arrow-up text-lg text-[#62666d] group-hover:text-[#8a8f98] transition-colors"></i>
+
+        <div class="text-center">
+          <p v-if="disabled" class="text-sm text-[#62666d]">
+            请先在 <span class="text-[rgb(145,132,235)]">系统设置</span> 中配置模型
+          </p>
+          <p v-else class="text-sm text-[#8a8f98]">
+            拖拽文件到此处或 <span class="text-[rgb(145,132,235)] cursor-pointer">浏览文件</span>
+          </p>
+          <p class="mt-1 text-xs text-[#62666d]">PDF, PNG, JPG</p>
+        </div>
+      </div>
+
+      <input
+        ref="fileInputEl"
+        type="file"
+        class="hidden"
+        accept=".pdf,.png,.jpg,.jpeg,.bmp,.tiff,.webp"
+        multiple
+        @change="onFileInput"
+      />
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/OcrPreview.vue b/frontend/src/components/workspace/OcrPreview.vue
new file mode 100644
index 00000000..91b3a208
--- /dev/null
+++ b/frontend/src/components/workspace/OcrPreview.vue
@@ -0,0 +1,162 @@
+<script setup>
+/**
+ * OcrPreview.vue
+ * OCR 结果预览 — 图片 + bbox 标注叠加
+ */
+import { ref, computed, onBeforeUnmount } from 'vue'
+import SplitLoading from './SplitLoading.vue'
+
+const props = defineProps({
+  pages: { type: Array, default: () => [] },
+  loading: { type: Boolean, default: false },
+  loadingText: { type: String, default: '正在执行 OCR 识别...' },
+  previewUrl: { type: String, default: '' },
+})
+
+// 按钮已移至 ContentPanel toolbar，emits 不再需要
+
+// 每张图片在 object-contain 下的实际渲染区域（含偏移）
+const imageSizes = ref({})
+const imgRef = ref(null)
+
+function calcImageSize(img, pageIndex) {
+  if (!img) return
+  const cw = img.clientWidth, ch = img.clientHeight
+  const nw = img.naturalWidth, nh = img.naturalHeight
+  if (!nw || !nh) return
+  const scale = Math.min(cw / nw, ch / nh)
+  const rw = nw * scale, rh = nh * scale
+  const ox = (cw - rw) / 2, oy = (ch - rh) / 2
+  imageSizes.value = { ...imageSizes.value, [pageIndex]: { w: rw, h: rh, ox, oy } }
+}
+
+function onImageLoad(pageIndex, event) {
+  calcImageSize(event.target, pageIndex)
+}
+
+// 浏览器缩放/窗口大小变化时重新计算
+const resizeObserver = new ResizeObserver(() => {
+  if (imgRef.value) {
+    calcImageSize(imgRef.value, currentPageData.value?.page_index)
+  }
+})
+onBeforeUnmount(() => resizeObserver.disconnect())
+
+// 颜色映射
+const labelColors = {
+  text: { border: 'rgba(59,130,246,0.6)', bg: 'rgba(59,130,246,0.15)', tag: 'rgba(59,130,246,0.85)' },
+  title: { border: 'rgba(168,85,247,0.6)', bg: 'rgba(168,85,247,0.15)', tag: 'rgba(168,85,247,0.85)' },
+  formula: { border: 'rgba(245,158,11,0.6)', bg: 'rgba(245,158,11,0.15)', tag: 'rgba(245,158,11,0.85)' },
+  image: { border: 'rgba(16,185,129,0.6)', bg: 'rgba(16,185,129,0.15)', tag: 'rgba(16,185,129,0.85)' },
+  chart: { border: 'rgba(16,185,129,0.6)', bg: 'rgba(16,185,129,0.15)', tag: 'rgba(16,185,129,0.85)' },
+  table: { border: 'rgba(236,72,153,0.6)', bg: 'rgba(236,72,153,0.15)', tag: 'rgba(236,72,153,0.85)' },
+  aside_text: { border: 'rgba(148,163,184,0.4)', bg: 'rgba(148,163,184,0.08)', tag: 'rgba(148,163,184,0.7)' },
+  header: { border: 'rgba(148,163,184,0.4)', bg: 'rgba(148,163,184,0.08)', tag: 'rgba(148,163,184,0.7)' },
+  footer: { border: 'rgba(148,163,184,0.4)', bg: 'rgba(148,163,184,0.08)', tag: 'rgba(148,163,184,0.7)' },
+}
+const defaultColor = { border: 'rgba(148,163,184,0.5)', bg: 'rgba(148,163,184,0.1)', tag: 'rgba(148,163,184,0.75)' }
+
+function getColor(label) {
+  return labelColors[label] || defaultColor
+}
+
+function boxStyle(block, page, pageIndex) {
+  const size = imageSizes.value[pageIndex]
+  if (!size) return { display: 'none' }
+  const scaleX = size.w / page.page_width
+  const scaleY = size.h / page.page_height
+  const [x1, y1, x2, y2] = block.bbox
+  const color = getColor(block.label)
+  return {
+    position: 'absolute',
+    left: `${size.ox + x1 * scaleX}px`,
+    top: `${size.oy + y1 * scaleY}px`,
+    width: `${(x2 - x1) * scaleX}px`,
+    height: `${(y2 - y1) * scaleY}px`,
+    border: `1.5px solid ${color.border}`,
+    backgroundColor: color.bg,
+    pointerEvents: 'none',
+  }
+}
+
+function tagStyle(block) {
+  const color = getColor(block.label)
+  return {
+    position: 'absolute',
+    top: '-1px',
+    left: '-1px',
+    backgroundColor: color.tag,
+    color: '#fff',
+    fontSize: '9px',
+    lineHeight: '1',
+    padding: '2px 4px',
+    borderRadius: '0 0 3px 0',
+    whiteSpace: 'nowrap',
+  }
+}
+
+// 分页
+const currentPage = ref(0)
+const currentPageData = computed(() => props.pages[currentPage.value])
+
+// 悬停的 block
+const hoveredBlock = ref(null)
+</script>
+
+<template>
+  <div class="flex flex-col h-full">
+    <!-- 加载中 -->
+    <div v-if="loading" class="flex-1 min-h-0 relative">
+      <img v-if="previewUrl" :src="previewUrl" class="absolute inset-0 w-full h-full object-contain opacity-10 blur-sm" alt="" />
+      <SplitLoading title="正在执行 OCR 识别" subtitle="PaddleOCR 正在解析文档结构与文字内容" />
+    </div>
+
+    <!-- 无数据 -->
+    <div v-else-if="!pages.length" class="flex-1 flex items-center justify-center">
+      <p class="text-sm text-[#62666d]">暂无 OCR 数据</p>
+    </div>
+
+    <!-- 预览（单张 + 分页） -->
+    <template v-else>
+      <div class="flex-1 min-h-0 relative overflow-hidden">
+        <img
+          v-if="currentPageData.image_url"
+          ref="imgRef"
+          :src="currentPageData.image_url"
+          class="block w-full h-full object-contain"
+          alt=""
+          @load="(e) => { onImageLoad(currentPageData.page_index, e); resizeObserver.disconnect(); resizeObserver.observe(e.target) }"
+        />
+        <!-- bbox 叠加层 -->
+        <div
+          v-for="(block, bi) in currentPageData.blocks"
+          :key="bi"
+          :style="boxStyle(block, currentPageData, currentPageData.page_index)"
+          @mouseenter="hoveredBlock = `${currentPageData.page_index}-${bi}`"
+          @mouseleave="hoveredBlock = null"
+          style="pointer-events: auto; cursor: default;"
+        >
+          <span :style="tagStyle(block)">{{ block.label }}</span>
+          <div
+            v-if="hoveredBlock === `${currentPageData.page_index}-${bi}` && block.content"
+            class="absolute left-0 top-full z-50 mt-1 max-w-xs rounded-md bg-slate-900/95 border border-white/10 px-2 py-1 text-xs text-slate-200 shadow-lg"
+            style="pointer-events: none;"
+          >
+            {{ block.content }}
+          </div>
+        </div>
+      </div>
+
+      <!-- 分页（多页时显示） -->
+      <div v-if="pages.length > 1" class="flex items-center justify-center gap-2 py-2 shrink-0">
+        <button
+          v-for="(_, i) in pages"
+          :key="i"
+          @click="currentPage = i"
+          class="h-2 rounded-full transition-all"
+          :class="i === currentPage ? 'w-6 bg-[rgb(129,115,223)]' : 'w-2 bg-white/20 hover:bg-white/40'"
+        />
+      </div>
+    </template>
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/ReviewStage.vue b/frontend/src/components/workspace/ReviewStage.vue
new file mode 100644
index 00000000..ab4369f0
--- /dev/null
+++ b/frontend/src/components/workspace/ReviewStage.vue
@@ -0,0 +1,73 @@
+<script setup>
+/**
+ * ReviewStage.vue
+ * 工作台第二页：分割结果核对（擦除预览 / OCR 预览 / 分割中 / 题目列表）
+ */
+import { ref } from 'vue'
+import ErasePreview from '@/components/workspace/ErasePreview.vue'
+import OcrPreview from '@/components/workspace/OcrPreview.vue'
+import SplitLoading from '@/components/workspace/SplitLoading.vue'
+import QuestionList from '@/components/question/QuestionList.vue'
+
+const props = defineProps({
+  eraseLoading: Boolean,
+  eraseDone: Boolean,
+  eraseImages: Array,
+  ocrLoading: Boolean,
+  ocrDone: Boolean,
+  ocrPages: Array,
+  splitting: Boolean,
+  splitCompleted: Boolean,
+  questions: Array,
+  selectedIds: Set,
+  previewUrl: String,
+})
+
+const emit = defineEmits([
+  'toggle-select', 'select-all', 'deselect-all', 'open-image', 'reorder',
+])
+
+const questionListRef = ref(null)
+
+defineExpose({
+  triggerTypeset: () => questionListRef.value?.triggerTypeset?.(),
+})
+</script>
+
+<template>
+  <!-- 擦除加载中 / 擦除对比预览 -->
+  <ErasePreview
+    v-if="eraseLoading || (eraseDone && !ocrLoading && !ocrDone)"
+    :images="eraseImages"
+    :loading="eraseLoading"
+    :preview-url="previewUrl"
+  />
+
+  <!-- OCR 加载中 / OCR 预览 -->
+  <OcrPreview
+    v-else-if="ocrLoading || (ocrDone && !splitCompleted && !splitting)"
+    :pages="ocrPages"
+    :loading="ocrLoading"
+    :preview-url="previewUrl"
+  />
+
+  <!-- 分割进行中 -->
+  <div v-else-if="splitting" class="flex-1 min-h-0 relative">
+    <img v-if="previewUrl" :src="previewUrl" class="absolute inset-0 w-full h-full object-contain opacity-10 blur-sm" alt="" />
+    <SplitLoading />
+  </div>
+
+  <!-- 题目列表 -->
+  <div v-if="splitCompleted && questions.length" class="flex-1 overflow-y-auto pr-2 custom-scrollbar py-2 pb-24">
+    <QuestionList
+      ref="questionListRef"
+      :questions="questions"
+      :selected-ids="selectedIds"
+      @toggle-select="(id) => emit('toggle-select', id)"
+      @select-all="emit('select-all')"
+      @deselect-all="emit('deselect-all')"
+      @open-image="(src) => emit('open-image', src)"
+      @reorder="(qs) => emit('reorder', qs)"
+    />
+  </div>
+</template>
diff --git a/frontend/src/components/SelectionPanel.vue b/frontend/src/components/workspace/SelectionPanel.vue
similarity index 95%
rename from frontend/src/components/SelectionPanel.vue
rename to frontend/src/components/workspace/SelectionPanel.vue
index 83ab9037..f512e616 100644
--- a/frontend/src/components/SelectionPanel.vue
+++ b/frontend/src/components/workspace/SelectionPanel.vue
@@ -1,4 +1,8 @@
 <script setup>
+/**
+ * SelectionPanel.vue
+ * 题目选择面板
+ */
 defineProps({
   count: { type: Number, default: 0 },
   visible: { type: Boolean, default: false },
@@ -19,7 +23,7 @@ const emit = defineEmits(['export', 'save', 'clear'])
     leave-to-class="translate-y-12"
   >
     <div v-if="visible && count" class="fixed bottom-24 left-1/2 z-50 -translate-x-1/2 md:bottom-10">
-      <div class="flex items-center gap-6 rounded-full border border-slate-200/60 bg-white/70 px-6 py-3 shadow-md backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]">
+      <div class="flex items-center gap-6 rounded-full border border-slate-200/60 bg-white/70 px-6 py-3 shadow-md dark:border-white/10 dark:bg-white/[0.03]">
         <div class="flex items-center gap-3 border-r border-slate-200/60 pr-6 dark:border-white/10">
           <div class="flex h-8 w-8 items-center justify-center rounded-full bg-blue-600 font-mono text-xs font-bold text-white shadow-sm dark:bg-indigo-500">
             {{ count }}
diff --git a/frontend/src/components/workspace/SidebarNav.vue b/frontend/src/components/workspace/SidebarNav.vue
new file mode 100644
index 00000000..4a21cbcc
--- /dev/null
+++ b/frontend/src/components/workspace/SidebarNav.vue
@@ -0,0 +1,313 @@
+<script setup>
+/**
+ * SidebarNav.vue
+ * 工作台左侧边栏导航（PC 端）+ 底部 Tab 导航（移动端）
+ */
+import BaseLogo from '@/components/base/BaseLogo.vue'
+
+const props = defineProps({
+  currentView: { type: String, required: true },
+  lastWorkspaceView: { type: String, default: 'workspace' },
+  currentUser: { type: Object, default: null },
+  isDark: { type: Boolean, default: true },
+  theme: { type: String, default: 'dark' },
+  // 导航配置
+  navGroups: { type: Array, required: true },
+  workspaceViews: { type: Object, required: true },
+  collapsedGroups: { type: Object, required: true },
+  // 指示器
+  navRef: { type: Object, default: null },
+  navBtnRefs: { type: Object, required: true },
+  indicatorStyle: { type: Object, default: () => ({}) },
+  indicatorTransition: { type: Boolean, default: false },
+  // AI 对话
+  chatCollapsed: { type: Boolean, default: false },
+  aiChatSessions: { type: Array, default: () => [] },
+  activeAiChatId: { type: [String, Number], default: null },
+  chatListRef: { type: Object, default: null },
+  chatBtnRefs: { type: Object, required: true },
+  chatIndicatorStyle: { type: Object, default: () => ({}) },
+  chatIndicatorTransition: { type: Boolean, default: false },
+  chatMenuOpenId: { type: [String, Number], default: null },
+  renamingChatId: { type: [String, Number], default: null },
+  renameText: { type: String, default: '' },
+  // 用户菜单
+  userMenuOpen: { type: Boolean, default: false },
+})
+
+const emit = defineEmits([
+  'update:currentView', 'update:collapsedGroups', 'update:chatCollapsed',
+  'update:userMenuOpen', 'update:chatMenuOpenId', 'update:renameText',
+  'navigate-home', 'logout', 'toggle-theme',
+  'create-ai-chat', 'select-ai-chat',
+  'start-rename-chat', 'confirm-rename-chat', 'delete-ai-chat', 'toggle-chat-menu',
+])
+
+const setView = (view) => emit('update:currentView', view)
+const toggleGroup = (gi) => {
+  const next = { ...props.collapsedGroups }
+  next[gi] = !next[gi]
+  emit('update:collapsedGroups', next)
+}
+</script>
+
+<template>
+  <!-- ================== PC端：左侧边栏导航 ================== -->
+  <aside class="hidden w-64 flex-col justify-between md:flex z-20">
+    <div>
+      <!-- Logo 标题区 -->
+      <div class="flex h-20 items-center justify-between px-4 py-6">
+        <button @click="emit('navigate-home')" class="flex min-w-0 items-center gap-2 rounded-md px-1 py-1 hover:bg-white/[0.04] transition-colors" title="返回首页">
+          <BaseLogo size="sm" />
+          <span class="text-sm font-medium text-[#f7f8f8]">智卷错题本</span>
+        </button>
+        <div class="flex items-center gap-1">
+          <button @click="setView('settings')" class="flex h-7 w-7 items-center justify-center rounded-md text-[#62666d] hover:bg-white/[0.04] hover:text-[#8a8f98] transition-colors" title="系统设置">
+            <i class="fa-solid fa-gear text-xs"></i>
+          </button>
+          <button @click="emit('logout')" class="flex h-7 w-7 items-center justify-center rounded-md border border-white/[0.08] text-[#62666d] hover:bg-white/[0.04] hover:text-[#8a8f98] transition-colors" title="退出登录">
+            <i class="fa-solid fa-right-from-bracket text-xs"></i>
+          </button>
+        </div>
+      </div>
+
+      <!-- 视图切换菜单 — Linear 分组折叠 -->
+      <nav :ref="(el) => $emit('update:navRef', el)" class="flex flex-col gap-1.5 px-4 relative">
+        <!-- 滑动指示器 -->
+        <div
+          class="absolute left-4 right-4 z-0 rounded-lg overflow-hidden brand-btn"
+          :class="indicatorTransition ? 'transition-all duration-300 ease-out' : ''"
+          :style="indicatorStyle"
+        ></div>
+
+        <template v-for="(group, gi) in navGroups" :key="gi">
+          <!-- 分组标题（可折叠） -->
+          <button
+            v-if="group.label"
+            @click="group.collapsible && toggleGroup(gi)"
+            class="flex items-center gap-1 px-3 pt-4 pb-1 text-xs font-medium uppercase tracking-[0.15em] text-[#62666d] hover:text-[#8a8f98] transition-colors"
+            :class="group.collapsible ? 'cursor-pointer' : 'cursor-default'"
+          >
+            <span>{{ group.label }}</span>
+            <i
+              v-if="group.collapsible"
+              class="fa-solid fa-play text-[8px] text-[#62666d] transition-transform duration-200"
+              :class="collapsedGroups[gi] ? '' : 'rotate-90'"
+            ></i>
+          </button>
+
+          <!-- 分组内容（grid 折叠动画） -->
+          <div class="grid transition-[grid-template-rows] duration-200 ease-out" :class="collapsedGroups[gi] ? 'grid-rows-[0fr]' : 'grid-rows-[1fr]'">
+          <div class="overflow-hidden">
+          <div class="flex flex-col gap-1">
+            <template v-for="item in group.items" :key="item.id">
+              <!-- 禁用项 -->
+              <button
+                v-if="item.disabled"
+                disabled
+                class="flex items-center justify-between rounded-lg px-3 py-3 text-sm cursor-not-allowed text-[#62666d]"
+              >
+                <div class="flex items-center gap-3">
+                  <i class="fa-solid w-4 text-center text-sm" :class="item.icon"></i>
+                  <span>{{ item.label }}</span>
+                </div>
+                <span class="text-[10px] font-medium px-2 py-0.5 rounded-md bg-white/[0.04] text-[#62666d]">敬请期待</span>
+              </button>
+              <!-- 普通项 -->
+              <button
+                v-else
+                :ref="el => navBtnRefs[item.id] = el"
+                @click="setView(item.id === 'workspace' ? lastWorkspaceView : item.id)"
+                class="group relative z-10 flex items-center gap-3 rounded-lg px-3 py-2 text-sm font-medium transition-colors duration-200"
+                :class="item.match(currentView)
+                  ? 'text-white'
+                  : 'text-[#8a8f98] hover:bg-white/[0.04] hover:text-[#d0d6e0]'"
+              >
+                <i class="fa-solid w-4 text-center text-sm" :class="item.icon"></i>
+                <span>{{ item.label }}</span>
+              </button>
+            </template>
+          </div>
+          </div>
+          </div>
+        </template>
+      </nav>
+    </div>
+
+    <!-- AI 对话历史列表 -->
+    <div class="flex-1 min-h-0 flex flex-col mt-4 px-4">
+      <div class="flex items-center justify-between px-3 pt-4 pb-2">
+        <button @click="emit('update:chatCollapsed', !chatCollapsed)" class="flex items-center gap-1 text-xs font-medium uppercase tracking-[0.15em] text-[#62666d] hover:text-[#8a8f98] transition-colors cursor-pointer">
+          <span>对话</span>
+          <i class="fa-solid fa-play text-[8px] text-[#62666d] transition-transform duration-200" :class="chatCollapsed ? '' : 'rotate-90'"></i>
+        </button>
+        <button @click="emit('create-ai-chat')" class="text-[#8a8f98] hover:text-[#d0d6e0] transition-colors">
+          <i class="fa-solid fa-plus text-[10px]"></i>
+        </button>
+      </div>
+      <!-- 折叠动画 -->
+      <div class="grid transition-[grid-template-rows] duration-200 ease-out" :class="chatCollapsed ? 'grid-rows-[0fr]' : 'grid-rows-[1fr]'">
+      <div class="overflow-hidden">
+      <div class="flex-1 overflow-y-auto pb-2 custom-scrollbar relative" @click="emit('update:chatMenuOpenId', null)">
+        <!-- 对话区滑动指示器 -->
+        <div
+          class="absolute left-0 right-0 z-0 rounded-md overflow-hidden brand-btn"
+          :class="chatIndicatorTransition ? 'transition-all duration-300 ease-out' : ''"
+          :style="chatIndicatorStyle"
+        ></div>
+
+        <div v-if="aiChatSessions.length === 0" class="px-3 py-4 text-center text-xs text-[#62666d]">
+          暂无对话
+        </div>
+        <div
+          v-for="s in aiChatSessions"
+          :key="s.id"
+          :ref="el => chatBtnRefs[s.id] = el"
+          class="group relative z-10 flex items-center gap-2 px-3 py-1.5 rounded-md mb-px cursor-pointer transition-colors"
+          :class="activeAiChatId === s.id && currentView === 'ai-chat'
+            ? 'text-white'
+            : 'text-[#8a8f98] hover:bg-white/[0.04] hover:text-[#d0d6e0]'"
+          @click="renamingChatId !== s.id && emit('select-ai-chat', s)"
+        >
+          <i class="fa-solid fa-message text-[10px] shrink-0" :class="activeAiChatId === s.id && currentView === 'ai-chat' ? 'text-white/60' : 'text-[#62666d]'"></i>
+
+          <!-- 重命名输入框 -->
+          <input
+            v-if="renamingChatId === s.id"
+            :value="renameText"
+            @input="emit('update:renameText', $event.target.value)"
+            data-rename-input
+            @click.stop
+            @keydown.enter="emit('confirm-rename-chat', s)"
+            @keydown.escape="$emit('update:renamingChatId', null)"
+            @blur="emit('confirm-rename-chat', s)"
+            class="flex-1 min-w-0 bg-transparent text-xs outline-none border-b border-white/[0.12] py-0.5 text-[#f7f8f8]"
+          />
+          <span v-else class="relative z-10 flex-1 truncate text-xs">{{ s.title }}</span>
+
+          <!-- 三个点按钮 -->
+          <button
+            @click.stop="emit('toggle-chat-menu', s.id)"
+            class="shrink-0 opacity-0 group-hover:opacity-100 text-[#62666d] hover:text-[#d0d6e0] transition-all"
+          >
+            <i class="fa-solid fa-ellipsis text-[10px]"></i>
+          </button>
+
+          <!-- Dropdown 菜单 -->
+          <Transition
+            enter-active-class="transition duration-100 ease-out"
+            enter-from-class="opacity-0 scale-95"
+            enter-to-class="opacity-100 scale-100"
+            leave-active-class="transition duration-75 ease-in"
+            leave-from-class="opacity-100 scale-100"
+            leave-to-class="opacity-0 scale-95"
+          >
+            <div
+              v-if="chatMenuOpenId === s.id"
+              class="absolute right-2 top-full mt-1 z-50 w-32 rounded-md brand-btn overflow-hidden"
+              @click.stop
+            >
+              <button
+                @click="emit('start-rename-chat', s)"
+                class="flex w-full items-center gap-2 px-3 py-1.5 text-xs text-[#d0d6e0] hover:bg-white/[0.05] transition-colors"
+              >
+                <i class="fa-solid fa-pen text-[10px] w-3 text-center text-[#62666d]"></i> 重命名
+              </button>
+              <button
+                @click="emit('update:chatMenuOpenId', null); emit('delete-ai-chat', s.id)"
+                class="flex w-full items-center gap-2 px-3 py-1.5 text-xs text-rose-400 hover:bg-rose-500/10 transition-colors"
+              >
+                <i class="fa-solid fa-trash text-[10px] w-4 text-center"></i> 删除
+              </button>
+            </div>
+          </Transition>
+        </div>
+      </div>
+      </div>
+      </div>
+    </div>
+
+    <!-- 底部用户区 -->
+    <div class="relative p-2">
+      <!-- Dropdown 菜单 -->
+      <Transition
+        enter-active-class="transition duration-150 ease-out"
+        enter-from-class="opacity-0 translate-y-2"
+        enter-to-class="opacity-100 translate-y-0"
+        leave-active-class="transition duration-100 ease-in"
+        leave-from-class="opacity-100 translate-y-0"
+        leave-to-class="opacity-0 translate-y-2"
+      >
+        <div v-if="userMenuOpen" class="absolute bottom-full left-2 right-2 mb-1 rounded-md brand-btn overflow-hidden z-50">
+          <button
+            @click="setView('settings'); emit('update:userMenuOpen', false)"
+            class="flex w-full items-center gap-2.5 px-3 py-2 text-sm text-[#d0d6e0] hover:bg-white/[0.05] transition-colors"
+          >
+            <i class="fa-solid fa-gear w-4 text-center text-xs text-[#62666d]"></i>
+            系统设置
+          </button>
+          <button
+            @click="(e) => { emit('update:userMenuOpen', false); emit('toggle-theme', e.currentTarget) }"
+            class="flex w-full items-center gap-2.5 px-3 py-2 text-sm text-[#d0d6e0] hover:bg-white/[0.05] transition-colors"
+          >
+            <i class="fa-solid w-4 text-center text-xs text-[#62666d]" :class="isDark ? 'fa-sun' : 'fa-moon'"></i>
+            {{ isDark ? '浅色模式' : '深色模式' }}
+          </button>
+          <div class="border-t border-white/[0.05]"></div>
+          <button
+            @click="emit('logout'); emit('update:userMenuOpen', false)"
+            class="flex w-full items-center gap-3 px-4 py-3 text-sm font-bold text-rose-500 hover:bg-rose-50 dark:hover:bg-rose-500/10 transition-colors"
+          >
+            <i class="fas fa-right-from-bracket w-4 text-center text-xs"></i>
+            退出登录
+          </button>
+        </div>
+      </Transition>
+
+      <!-- 用户信息 -->
+      <button
+        @click="emit('update:userMenuOpen', !userMenuOpen)"
+        class="flex w-full items-center gap-2 px-2 py-1.5 rounded-md hover:bg-white/[0.04] transition-colors"
+      >
+        <div class="h-8 w-8 shrink-0 rounded-xl relative overflow-hidden flex items-center justify-center text-white text-sm font-medium" style="background: linear-gradient(to bottom, rgba(129,115,223,0.9), rgba(99,87,199,0.9)); box-shadow: inset 0 1px 0 0 rgba(255,255,255,0.12);">
+          <span class="absolute inset-0 pointer-events-none" style="background-image: linear-gradient(to right, rgba(255,255,255,0.06) 1px, transparent 1px), linear-gradient(to bottom, rgba(255,255,255,0.06) 1px, transparent 1px); background-size: 8px 8px; mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%); -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);"></span>
+          <span class="relative z-10">{{ currentUser?.username?.[0]?.toUpperCase() ?? '?' }}</span>
+        </div>
+        <div class="flex-1 min-w-0 text-left">
+          <p class="text-sm text-[#f7f8f8] truncate leading-tight">{{ currentUser?.username }}</p>
+        </div>
+        <i class="fa-solid fa-chevron-up text-[10px] text-[#62666d]"></i>
+      </button>
+    </div>
+  </aside>
+
+  <!-- ================== 移动端：底部 Tab 导航栏 ================== -->
+  <nav class="fixed bottom-0 left-0 right-0 z-50 border-t border-white/[0.06] bg-[#0A0A0F]/90 pb-2 pt-2 md:hidden">
+    <div class="flex justify-around">
+      <button @click="setView(lastWorkspaceView)" class="flex flex-col items-center p-2" :class="workspaceViews.has(currentView) ? 'text-indigo-400' : 'text-white/40'">
+        <i class="fa-solid fa-file-arrow-up text-lg"></i>
+        <span class="mt-1 text-xs font-bold">录入</span>
+      </button>
+      <button @click="setView('notes')" class="flex flex-col items-center p-2" :class="currentView === 'notes' ? 'text-indigo-400' : 'text-white/40'">
+        <i class="fa-solid fa-book-open text-lg"></i>
+        <span class="mt-1 text-xs font-bold">笔记库</span>
+      </button>
+      <button @click="setView('dashboard')" class="flex flex-col items-center p-2" :class="currentView === 'dashboard' ? 'text-indigo-400' : 'text-white/40'">
+        <i class="fa-solid fa-chart-pie text-lg"></i>
+        <span class="mt-1 text-xs font-bold">数据面板</span>
+      </button>
+      <button @click="setView('error-bank')" class="flex flex-col items-center p-2" :class="currentView === 'error-bank' ? 'text-indigo-400' : 'text-white/40'">
+        <i class="fa-solid fa-layer-group text-lg"></i>
+        <span class="mt-1 text-xs font-bold">错题本</span>
+      </button>
+      <button @click="setView('settings')" class="flex flex-col items-center p-2" :class="currentView === 'settings' ? 'text-indigo-400' : 'text-white/40'">
+        <i class="fa-solid fa-sliders text-lg"></i>
+        <span class="mt-1 text-xs font-bold">设置</span>
+      </button>
+      <button @click="(e) => emit('toggle-theme', e.currentTarget)" class="flex flex-col items-center p-2 text-white/40">
+        <i class="fa-solid text-lg" :class="theme === 'dark' ? 'fa-sun' : 'fa-moon'"></i>
+        <span class="mt-1 text-xs font-bold">主题</span>
+      </button>
+    </div>
+  </nav>
+</template>
diff --git a/frontend/src/components/SplitLoading.vue b/frontend/src/components/workspace/SplitLoading.vue
similarity index 76%
rename from frontend/src/components/SplitLoading.vue
rename to frontend/src/components/workspace/SplitLoading.vue
index aa1d58a3..c3720b04 100644
--- a/frontend/src/components/SplitLoading.vue
+++ b/frontend/src/components/workspace/SplitLoading.vue
@@ -1,5 +1,14 @@
 <script setup>
-// AI 现代化扫描处理动画
+/**
+ * SplitLoading.vue
+ * 分割加载动画
+ */
+import BaseLogo from '@/components/base/BaseLogo.vue'
+
+defineProps({
+  title: { type: String, default: 'AI 正在深度解析' },
+  subtitle: { type: String, default: '正在识别题目结构、提取知识点并构建图谱' },
+})
 </script>
 
 <template>
@@ -18,7 +27,7 @@
 
           <!-- 核心图标 -->
           <div class="core-icon">
-            <i class="fa-solid fa-brain"></i>
+            <BaseLogo size="lg" />
             <!-- 核心脉冲 -->
             <div class="core-pulse"></div>
           </div>
@@ -29,8 +38,8 @@
 
         <!-- 文本引导 -->
         <div class="text-container">
-          <h3 class="status-title">AI 正在深度解析</h3>
-          <p class="status-subtitle">正在识别题目结构、提取知识点并构建图谱</p>
+          <h3 class="status-title">{{ title }}</h3>
+          <p class="status-subtitle">{{ subtitle }}</p>
 
           <!-- 模拟进度条 -->
           <div class="progress-container">
@@ -52,12 +61,7 @@
   display: flex;
   align-items: center;
   justify-content: center;
-  background: rgba(255, 255, 255, 0.6);
-  backdrop-filter: blur(20px) saturate(180%);
-}
-
-:root.dark .loading-overlay {
-  background: rgba(10, 10, 15, 0.8);
+  background: transparent;
 }
 
 .loading-content {
@@ -81,7 +85,7 @@
   position: absolute;
   width: 120%;
   height: 120%;
-  background: radial-gradient(circle, rgba(59, 130, 246, 0.15) 0%, transparent 70%);
+  background: radial-gradient(circle, rgba(129, 115, 223, 0.15) 0%, transparent 70%);
   filter: blur(20px);
   animation: breathe 4s ease-in-out infinite;
 }
@@ -90,8 +94,8 @@
 .orbit {
   position: absolute;
   border-radius: 50%;
-  border: 1.5px solid rgba(59, 130, 246, 0.1);
-  border-top-color: rgba(59, 130, 246, 0.5);
+  border: 1.5px solid rgba(129, 115, 223, 0.1);
+  border-top-color: rgba(129, 115, 223, 0.5);
 }
 
 .orbit-1 {
@@ -104,43 +108,30 @@
   width: 75%;
   height: 75%;
   animation: rotate 5s linear infinite reverse;
-  border-top-color: rgba(99, 102, 241, 0.5);
+  border-top-color: rgba(145, 132, 235, 0.5);
 }
 
 .orbit-3 {
   width: 50%;
   height: 50%;
   animation: rotate 3s linear infinite;
-  border-top-color: rgba(139, 92, 246, 0.5);
+  border-top-color: rgba(129, 115, 223, 0.5);
 }
 
 /* 核心图标 */
 .core-icon {
   position: relative;
   z-index: 10;
-  width: 56px;
-  height: 56px;
-  background: white;
-  border-radius: 1rem;
   display: flex;
   align-items: center;
   justify-content: center;
-  box-shadow: 0 10px 25px rgba(59, 130, 246, 0.2);
-  color: #2563eb;
-  font-size: 1.5rem;
-}
-
-:root.dark .core-icon {
-  background: #1e1e2e;
-  box-shadow: 0 10px 25px rgba(0, 0, 0, 0.3);
-  color: #818cf8;
 }
 
 .core-pulse {
   position: absolute;
   inset: -4px;
   border-radius: 1.25rem;
-  background: rgba(59, 130, 246, 0.2);
+  background: rgba(129, 115, 223, 0.2);
   animation: pulse 2s ease-out infinite;
 }
 
@@ -150,8 +141,8 @@
   left: 0;
   right: 0;
   height: 2px;
-  background: linear-gradient(90deg, transparent, rgba(59, 130, 246, 0.8), transparent);
-  box-shadow: 0 0 15px rgba(59, 130, 246, 0.5);
+  background: linear-gradient(90deg, transparent, rgba(129, 115, 223, 0.8), transparent);
+  box-shadow: 0 0 15px rgba(129, 115, 223, 0.5);
   z-index: 5;
   animation: scan 3s ease-in-out infinite;
 }
@@ -186,7 +177,7 @@
   margin-top: 1.5rem;
   width: 100%;
   height: 4px;
-  background: rgba(59, 130, 246, 0.05);
+  background: rgba(129, 115, 223, 0.05);
   border-radius: 10px;
   overflow: hidden;
 }
@@ -194,7 +185,7 @@
 .progress-bar {
   width: 40%;
   height: 100%;
-  background: linear-gradient(90deg, #2563eb, #818cf8);
+  background: linear-gradient(90deg, rgb(129,115,223), rgb(145,132,235));
   border-radius: 10px;
   position: relative;
   animation: progress-move 2.5s ease-in-out infinite;
diff --git a/frontend/src/components/StatusBar.vue b/frontend/src/components/workspace/StatusBar.vue
similarity index 67%
rename from frontend/src/components/StatusBar.vue
rename to frontend/src/components/workspace/StatusBar.vue
index b6e682ab..f9e89d56 100644
--- a/frontend/src/components/StatusBar.vue
+++ b/frontend/src/components/workspace/StatusBar.vue
@@ -1,16 +1,44 @@
 <script setup>
-import { computed, ref, watch, inject } from 'vue'
+/**
+ * StatusBar.vue
+ * 状态栏（引擎状态 + 模型选择）
+ */
+import { computed, ref, watch } from 'vue'
+import { useToast } from '@/composables/useToast.js'
 import {
   Listbox,
   ListboxButton,
   ListboxOptions,
   ListboxOption,
 } from '@headlessui/vue'
-import deepseekLogo from '../assets/deepseek.svg'
-import ernieLogo from '../assets/ernie.svg'
+import deepseekLogo from '@/assets/deepseek.svg'
+import ernieLogo from '@/assets/ernie.svg'
+
+// Tooltip 相关状态
+const tooltipVisible = ref(false)
+const tooltipText = ref('')
+const tooltipPosition = ref({ x: 0, y: 0 })
+
+// 显示 tooltip
+const showTooltip = (event, text) => {
+  if (!text || text.length <= 20) return // 短文本不需要 tooltip
+
+  tooltipText.value = text
+  const rect = event.currentTarget.getBoundingClientRect()
+  tooltipPosition.value = {
+    x: rect.left + rect.width / 2,
+    y: rect.top - 8
+  }
+  tooltipVisible.value = true
+}
+
+// 隐藏 tooltip
+const hideTooltip = () => {
+  tooltipVisible.value = false
+}
 
 const emit = defineEmits(['update:selectedModel'])
-const pushToast = inject('pushToast')
+const { pushToast } = useToast()
 
 const props = defineProps({
   statusLoading: { type: Boolean, default: true },
@@ -93,18 +121,18 @@ const modelStatusError = computed(() => {
     class="relative z-20 flex flex-wrap items-center gap-4 py-2 text-sm shrink-0"
   >
     <div class="flex items-center gap-2.5">
-      <span class="text-[11px] font-bold uppercase tracking-[0.15em] text-slate-600 dark:text-slate-400">
+      <span class="text-xs font-medium uppercase tracking-[0.15em] text-[#62666d]">
         引擎状态
       </span>
     </div>
 
     <!-- 分隔符 -->
-    <div class="h-4 w-px bg-slate-200 dark:bg-white/10"></div>
+    <div class="h-4 w-px bg-white/[0.08]"></div>
 
     <!-- 全局系统错误 -->
     <span
       v-if="statusError"
-      class="inline-flex items-center gap-2 rounded-lg border border-rose-200 bg-rose-50/80 px-3 py-1.5 text-[11px] font-bold text-rose-700 shadow-sm dark:border-rose-500/30 dark:bg-rose-500/10 dark:text-rose-400"
+      class="inline-flex items-center gap-2 rounded-lg border border-rose-500/30 bg-rose-500/10 px-3 py-1.5 text-xs font-medium text-rose-400"
     >
       <i class="fa-solid fa-circle-exclamation animate-pulse"></i>
       {{ statusError }}
@@ -115,15 +143,15 @@ const modelStatusError = computed(() => {
       <span
         v-for="p in statusPills"
         :key="p.key"
-        class="inline-flex items-center gap-2 rounded-lg border px-3 py-1.5 text-[11px] font-bold"
+        class="inline-flex items-center gap-2 rounded-lg border px-3 py-1.5 text-xs font-medium"
         :class="
           p.loading
-            ? 'border-amber-200 bg-amber-50/80 text-amber-700 dark:border-amber-500/20 dark:bg-amber-500/10 dark:text-amber-400'
+            ? 'border-amber-500/20 bg-amber-500/10 text-amber-400'
             : p.isPlaceholder
-              ? 'border-slate-200 bg-slate-50/50 text-slate-400 dark:border-white/5 dark:bg-white/5 dark:text-slate-500'
+              ? 'border-white/[0.05] bg-white/[0.03] text-[#62666d]'
               : p.ok
-                ? 'border-emerald-200 bg-emerald-50/80 text-emerald-700 dark:border-emerald-500/20 dark:bg-emerald-500/10 dark:text-emerald-400 shadow-sm shadow-emerald-500/5'
-                : 'border-rose-200 bg-rose-50/80 text-rose-700 dark:border-rose-500/20 dark:bg-rose-500/10 dark:text-rose-400'
+                ? 'border-emerald-500/20 bg-emerald-500/10 text-emerald-400'
+                : 'border-rose-500/20 bg-rose-500/10 text-rose-400'
         "
       >
         <div class="relative h-2.5 w-2.5 shrink-0">
@@ -142,13 +170,13 @@ const modelStatusError = computed(() => {
     <!-- 模型下拉选择器 (按 provider 分组) -->
     <div v-if="!statusError" class="ml-auto flex items-center gap-2">
       <Listbox :model-value="selectedModel" @update:model-value="(v) => emit('update:selectedModel', v)" :disabled="disabled || noModels">
-        <div class="relative w-48">
+        <div class="relative w-56 min-w-0">
           <ListboxButton
-            class="group relative flex h-9 w-full cursor-pointer items-center justify-between gap-4 rounded-xl border border-slate-300 bg-white pl-3 pr-2 text-left shadow-sm backdrop-blur-sm hover:border-blue-400 hover:bg-slate-50 dark:border-white/5 dark:bg-white/5 dark:hover:border-indigo-500/20 dark:hover:bg-white/10"
+            class="group relative flex w-full cursor-pointer items-center justify-between gap-4 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-left text-xs font-medium text-[#d0d6e0] transition-colors hover:bg-white/[0.05] hover:border-white/[0.12]"
             :disabled="disabled"
           >
-            <div class="flex items-center gap-2.5">
-              <div class="relative flex h-6 w-6 shrink-0 items-center justify-center rounded-lg bg-slate-100 text-blue-600 dark:bg-white/5 dark:text-indigo-400">
+            <div class="flex items-center gap-2.5 min-w-0 flex-1">
+              <div class="relative flex h-6 w-6 shrink-0 items-center justify-center rounded-lg bg-white/[0.05] text-[rgb(145,132,235)]">
                 <Transition name="fade" mode="out-in">
                   <img
                     v-if="currentProvider && modelLogos[currentProvider.value]"
@@ -162,11 +190,15 @@ const modelStatusError = computed(() => {
                 <!-- 状态指示点 -->
                 <div
                   v-if="currentProvider || statusLoading"
-                  class="absolute -right-0.5 -top-0.5 h-2 w-2 rounded-full border border-white dark:border-[#0A0A0F]"
+                  class="absolute -right-0.5 -top-0.5 h-2 w-2 rounded-full border border-[#0f1011]"
                   :class="(statusLoading || isChecking) ? 'bg-amber-400 animate-pulse' : (!currentProvider.configured || currentProvider.key_valid === false) ? 'bg-rose-500' : 'bg-emerald-500'"
                 ></div>
               </div>
-              <span class="block max-w-[180px] truncate text-[11px] font-black tracking-tight text-slate-700 dark:text-slate-200">
+              <span
+                class="block min-w-0 flex-1 truncate text-xs font-medium tracking-tight text-[#d0d6e0]"
+                @mouseenter="showTooltip($event, selectedLabel)"
+                @mouseleave="hideTooltip"
+              >
                 {{ selectedLabel }}
               </span>
             </div>
@@ -178,7 +210,7 @@ const modelStatusError = computed(() => {
                   <i v-else-if="currentProvider" key="ok" class="fa-solid fa-check absolute inset-0 flex items-center justify-center text-[10px] text-emerald-500"></i>
                 </Transition>
               </div>
-              <i class="fa-solid fa-chevron-down shrink-0 text-[10px] text-slate-400 transition-transform duration-300 group-aria-expanded:rotate-180"></i>
+              <i class="fa-solid fa-chevron-down shrink-0 text-[10px] text-[#62666d] transition-transform duration-300 group-aria-expanded:rotate-180"></i>
             </div>
           </ListboxButton>
 
@@ -191,14 +223,14 @@ const modelStatusError = computed(() => {
             leave-to-class="transform scale-95 opacity-0 -translate-y-2"
           >
             <ListboxOptions
-              class="absolute right-0 z-50 mt-2 max-h-72 w-64 overflow-auto rounded-xl border border-slate-200/80 bg-white/95 py-2 text-base shadow-xl backdrop-blur-xl focus:outline-none sm:text-sm dark:border-white/10 dark:bg-[#0A0A0F]/95"
+              class="absolute right-0 z-50 mt-2 max-h-72 w-full overflow-auto rounded-md border border-white/[0.08] bg-white/[0.02] text-base focus:outline-none sm:text-sm"
             >
               <template v-for="(item, idx) in modelOptions" :key="idx">
                 <!-- 分组标题 -->
                 <li
                   v-if="item.type === 'group'"
-                  class="select-none px-4 py-2 text-[10px] font-bold uppercase tracking-widest text-slate-400 dark:text-slate-500"
-                  :class="idx > 0 ? 'mt-1 border-t border-slate-100 pt-2 dark:border-white/5' : ''"
+                  class="select-none px-4 text-[10px] font-medium uppercase tracking-widest text-[#62666d]"
+                  :class="idx > 0 ? 'mt-1 border-t border-white/[0.05] pt-2' : ''"
                 >
                   <div class="flex items-center gap-2">
                     <img v-if="modelLogos[item.provider]" :src="modelLogos[item.provider]" class="h-3 w-3 object-contain opacity-60" alt="" />
@@ -216,20 +248,22 @@ const modelStatusError = computed(() => {
                   <li
                     class="relative cursor-pointer select-none py-2.5 pl-10 pr-4 transition-colors"
                     :class="[
-                      active ? 'bg-blue-50 dark:bg-indigo-500/10' : '',
+                      active ? 'bg-white/[0.05]' : '',
                       optDisabled ? 'opacity-40 cursor-not-allowed' : ''
                     ]"
                   >
                     <span
-                      class="block truncate text-sm transition-colors"
-                      :class="[selected ? 'font-bold text-blue-700 dark:text-indigo-300' : 'font-medium text-slate-700 dark:text-slate-300']"
-                    >
-                      {{ item.model }}
-                      <span v-if="item.isDefault" class="ml-1 text-[10px] text-slate-400 dark:text-slate-500">默认</span>
-                    </span>
+                  class="block truncate text-sm transition-colors min-w-0"
+                  :class="[selected ? 'font-medium text-white' : 'font-medium text-[#d0d6e0]']"
+                  @mouseenter="showTooltip($event, item.model)"
+                  @mouseleave="hideTooltip"
+                >
+                  {{ item.model }}
+                  <span v-if="item.isDefault" class="ml-1 text-[10px] text-[#62666d]">默认</span>
+                </span>
                     <span
                       v-if="selected"
-                      class="absolute inset-y-0 left-0 flex items-center pl-3 text-blue-600 dark:text-indigo-400"
+                      class="absolute inset-y-0 left-0 flex items-center pl-3 text-white"
                     >
                       <i class="fa-solid fa-check text-sm"></i>
                     </span>
@@ -242,6 +276,40 @@ const modelStatusError = computed(() => {
       </Listbox>
     </div>
   </div>
+
+  <!-- Tooltip 组件 -->
+  <Teleport to="body">
+    <Transition
+      enter-active-class="transition duration-200 ease-out"
+      enter-from-class="opacity-0 scale-95 translate-y-1"
+      enter-to-class="opacity-100 scale-100 translate-y-0"
+      leave-active-class="transition duration-150 ease-in"
+      leave-from-class="opacity-100 scale-100 translate-y-0"
+      leave-to-class="opacity-0 scale-95 translate-y-1"
+    >
+      <div
+        v-if="tooltipVisible && tooltipText"
+        class="fixed z-[9999] pointer-events-none"
+        :style="{
+          left: `${tooltipPosition.x}px`,
+          top: `${tooltipPosition.y}px`,
+          transform: 'translateX(-50%) translateY(-100%)'
+        }"
+      >
+        <div class="relative">
+          <!-- 箭头 -->
+          <div class="absolute left-1/2 top-full -translate-x-1/2 w-2 h-2">
+            <div class="w-2 h-2 bg-[#191a1b] rotate-45"></div>
+          </div>
+
+          <!-- 内容 -->
+          <div class="bg-[#191a1b] text-[#d0d6e0] px-3 py-2 rounded-lg text-xs font-medium whitespace-nowrap">
+            {{ tooltipText }}
+          </div>
+        </div>
+      </div>
+    </Transition>
+  </Teleport>
 </template>
 
 <style scoped>
diff --git a/frontend/src/components/workspace/StepIndicator.vue b/frontend/src/components/workspace/StepIndicator.vue
new file mode 100644
index 00000000..ebf0c854
--- /dev/null
+++ b/frontend/src/components/workspace/StepIndicator.vue
@@ -0,0 +1,96 @@
+<script setup>
+/**
+ * StepIndicator.vue
+ * 步骤指示器
+ */
+import { computed } from 'vue'
+
+const props = defineProps({
+  step: { type: Number, default: 1 },
+  mode: { type: String, default: 'exam' }, // exam | note
+})
+
+const examLabels = ['上传文件', 'OCR解析', '分割题目', '预览导出']
+const examDescs = ['选择 PDF 或图片', '结构化提取题干', 'AI 识别并拆分', '勾选题目并导出']
+const noteLabels = ['上传笔记', 'OCR解析', '整理归纳', '保存笔记']
+const noteDescs = ['选择 PDF 或图片', '结构化提取内容', 'AI 智能整理', '存入笔记库']
+
+const labels = computed(() => props.mode === 'note' ? noteLabels : examLabels)
+const descs = computed(() => props.mode === 'note' ? noteDescs : examDescs)
+</script>
+
+<template>
+  <div class="px-2 sm:px-0">
+    <ol class="relative flex flex-col gap-6 sm:flex-row sm:items-start sm:justify-between sm:gap-0">
+      <li
+        v-for="n in 4"
+        :key="n"
+        class="group relative flex items-center gap-4 sm:flex-1 sm:flex-col sm:gap-2"
+      >
+        <!-- 桌面端连接线 (优化定位，避免穿过圆圈) -->
+        <div
+          v-if="n < 4"
+          class="hidden sm:block absolute left-[calc(50%+1.25rem)] top-5 w-[calc(100%-2.5rem)] h-[2px] -translate-y-1/2"
+          :class="n < step ? 'bg-[#62666d]' : 'bg-white/[0.08]'"
+        ></div>
+
+        <!-- 移动端连接线 -->
+        <div
+          v-if="n < 4"
+          class="sm:hidden absolute left-5 top-12 bottom-[-1.5rem] w-[2px] -translate-x-1/2"
+          :class="n < step ? 'bg-[#62666d]' : 'bg-white/[0.08]'"
+        ></div>
+
+        <!-- 步骤圆圈 -->
+        <div
+          class="step-circle relative z-10 flex h-10 w-10 shrink-0 items-center justify-center rounded-lg border"
+          :class="
+            n < step
+              ? 'border-transparent bg-gradient-to-br from-[rgb(129,115,223)] to-[rgb(145,132,235)] text-white/90'
+              : n === step
+                ? 'border-white/[0.12] bg-white/[0.08] text-[rgb(145,132,235)]'
+                : 'border-white/[0.05] bg-white/[0.03] text-[#62666d]'
+          "
+        >
+          <Transition name="scale" mode="out-in">
+            <i v-if="n < step" :key="'check-'+n" class="fa-solid fa-check text-[10px] font-medium"></i>
+            <span v-else :key="'num-'+n" class="text-xs font-medium tabular-nums">{{ n }}</span>
+          </Transition>
+
+          <!-- 当前步骤的光晕 -->
+          <div v-if="n === step" class="absolute -inset-1 animate-pulse rounded-lg bg-[rgb(129,115,223)]/20 blur-md"></div>
+        </div>
+
+        <!-- 文本描述 -->
+        <div class="flex flex-col sm:items-center sm:text-center">
+          <h3
+            class="text-xs font-medium tracking-tight"
+            :class="n <= step ? 'text-[#f7f8f8]' : 'text-[#62666d]'"
+          >
+            {{ labels[n - 1] }}
+          </h3>
+          <p class="mt-0.5 text-xs font-medium leading-relaxed text-[#62666d] sm:max-w-[100px]">
+            {{ descs[n - 1] }}
+          </p>
+        </div>
+      </li>
+    </ol>
+  </div>
+</template>
+
+<style scoped>
+.scale-enter-active,
+.scale-leave-active {
+  transition: all 0.3s cubic-bezier(0.34, 1.56, 0.64, 1);
+}
+
+.scale-enter-from {
+  opacity: 0;
+  transform: scale(0.5);
+}
+
+.scale-leave-to {
+  opacity: 0;
+  transform: scale(1.5);
+}
+</style>
diff --git a/frontend/src/components/workspace/UploadStage.vue b/frontend/src/components/workspace/UploadStage.vue
new file mode 100644
index 00000000..2fec2dae
--- /dev/null
+++ b/frontend/src/components/workspace/UploadStage.vue
@@ -0,0 +1,179 @@
+<script setup>
+/**
+ * UploadStage.vue
+ * 工作台第一页：上传与分析（模式切换 + 擦除开关 + 流程步骤 + 文件上传）
+ */
+import StatusBar from '@/components/workspace/StatusBar.vue'
+import FileUploader from '@/components/workspace/FileUploader.vue'
+import FileList from '@/components/workspace/FileList.vue'
+import ActionBar from '@/components/workspace/ActionBar.vue'
+
+const props = defineProps({
+  uploadMode: String,
+  eraseEnabled: Boolean,
+  // StatusBar
+  statusLoading: Boolean,
+  statusError: [String, null],
+  statusPills: Array,
+  providerOptions: Array,
+  selectedModel: String,
+  hasConfiguredModel: Boolean,
+  splitting: Boolean,
+  splitCompleted: Boolean,
+  // FileUploader / FileList
+  pendingFiles: Array,
+  fileProgress: Object,
+  waitingKeys: Object,
+  uploadBusy: Boolean,
+  uploadReady: Boolean,
+  // ActionBar
+  splitEnabled: Boolean,
+})
+
+const emit = defineEmits([
+  'update:upload-mode',
+  'update:erase-enabled',
+  'update:selected-model',
+  'upload',
+  'remove-file',
+  'split',
+])
+</script>
+
+<template>
+  <!-- 工具栏：状态 + 模式切换 + 擦除开关 -->
+  <div class="flex flex-wrap items-center gap-3 py-2">
+    <!-- 模式切换 -->
+    <div class="flex items-center rounded-md brand-btn p-0.5">
+      <button
+        @click="emit('update:upload-mode', 'exam')"
+        class="h-7 px-3 rounded text-xs font-medium transition-all"
+        :class="uploadMode === 'exam' ? 'bg-white/[0.06] text-[#f7f8f8]' : 'text-[#62666d] hover:text-[#8a8f98]'"
+      >
+        <i class="fa-solid fa-file-lines mr-1.5"></i>试卷分割
+      </button>
+      <button
+        @click="emit('update:upload-mode', 'note')"
+        class="h-7 px-3 rounded text-xs font-medium transition-all"
+        :class="uploadMode === 'note' ? 'bg-white/[0.06] text-[#f7f8f8]' : 'text-[#62666d] hover:text-[#8a8f98]'"
+      >
+        <i class="fa-solid fa-book-open mr-1.5"></i>笔记整理
+      </button>
+    </div>
+
+    <div class="h-4 w-px bg-white/[0.08]"></div>
+
+    <!-- 擦除开关 -->
+    <label class="flex cursor-pointer items-center gap-2" @click="emit('update:erase-enabled', !eraseEnabled)">
+      <div class="relative h-4 w-7 rounded-full transition-colors" :class="eraseEnabled ? 'bg-[rgb(129,115,223)]' : 'bg-white/[0.08]'">
+        <div class="absolute top-0.5 h-3 w-3 rounded-full bg-white transition-transform" :class="eraseEnabled ? 'translate-x-3' : 'translate-x-0.5'"></div>
+      </div>
+      <span class="text-xs text-[#8a8f98]">擦除笔迹</span>
+      <span class="relative group/tip">
+        <i class="fa-solid fa-circle-question text-[10px] text-[#62666d] cursor-help"></i>
+        <span class="absolute bottom-full right-0 mb-2 px-3 py-1.5 rounded-md bg-[#191a1b] border border-white/[0.08] text-xs text-[#d0d6e0] whitespace-nowrap opacity-0 pointer-events-none group-hover/tip:opacity-100 transition-opacity">
+          上传后自动擦除图片中的手写笔迹
+        </span>
+      </span>
+    </label>
+
+    <!-- 引擎状态 -->
+    <div class="ml-auto">
+      <StatusBar
+        :status-loading="statusLoading"
+        :status-error="statusError"
+        :status-pills="statusPills"
+        :provider-options="providerOptions"
+        :selected-model="selectedModel"
+        :disabled="splitting || splitCompleted"
+        :no-models="!hasConfiguredModel"
+        @update:selected-model="(v) => emit('update:selected-model', v)"
+      />
+    </div>
+  </div>
+
+  <!-- 上传区 -->
+  <div class="flex-1 min-h-0 overflow-y-auto custom-scrollbar flex flex-col items-center justify-center py-8 gap-6">
+    <!-- 引导信息 -->
+    <div class="text-center max-w-md">
+      <h3 class="text-base font-medium text-[#f7f8f8] mb-2">
+        {{ uploadMode === 'note' ? '上传手写笔记' : '上传试卷图片' }}
+      </h3>
+      <p class="text-sm text-[#62666d] leading-relaxed">
+        {{ uploadMode === 'note'
+          ? '支持拍照或扫描件，AI 将自动识别内容并整理为结构化笔记'
+          : '支持 PDF 和图片格式，AI 将自动完成 OCR 识别、题目分割和知识点标注'
+        }}
+      </p>
+    </div>
+
+    <!-- 流程步骤卡片 -->
+    <div class="grid grid-cols-4 gap-4 w-full max-w-2xl">
+      <div class="flex flex-col items-center gap-3 rounded-lg brand-btn p-4 text-center">
+        <div class="flex h-12 w-12 items-center justify-center rounded-lg bg-white/[0.04]">
+          <i class="fa-solid fa-cloud-arrow-up text-xl text-[rgb(129,115,223)]"></i>
+        </div>
+        <span class="text-sm text-[#8a8f98]">{{ uploadMode === 'note' ? '上传笔记' : '上传文件' }}</span>
+      </div>
+      <div class="flex flex-col items-center gap-3 rounded-lg brand-btn p-4 text-center">
+        <div class="flex h-12 w-12 items-center justify-center rounded-lg bg-white/[0.04]">
+          <i class="fa-solid fa-eye text-xl text-[rgb(129,115,223)]"></i>
+        </div>
+        <span class="text-sm text-[#8a8f98]">AI 识别</span>
+      </div>
+      <div class="flex flex-col items-center gap-3 rounded-lg brand-btn p-4 text-center">
+        <div class="flex h-12 w-12 items-center justify-center rounded-lg bg-white/[0.04]">
+          <i class="fa-solid text-xl text-[rgb(129,115,223)]" :class="uploadMode === 'note' ? 'fa-wand-magic-sparkles' : 'fa-scissors'"></i>
+        </div>
+        <span class="text-sm text-[#8a8f98]">{{ uploadMode === 'note' ? '智能整理' : '分割纠错' }}</span>
+      </div>
+      <div class="flex flex-col items-center gap-3 rounded-lg brand-btn p-4 text-center">
+        <div class="flex h-12 w-12 items-center justify-center rounded-lg bg-white/[0.04]">
+          <i class="fa-solid text-xl text-[rgb(129,115,223)]" :class="uploadMode === 'note' ? 'fa-bookmark' : 'fa-file-export'"></i>
+        </div>
+        <span class="text-sm text-[#8a8f98]">{{ uploadMode === 'note' ? '保存笔记' : '导出归档' }}</span>
+      </div>
+    </div>
+
+    <!-- 拖拽上传 -->
+    <FileUploader
+      :pending-files="pendingFiles"
+      :file-progress="fileProgress"
+      :waiting-keys="waitingKeys"
+      :upload-busy="uploadBusy"
+      :upload-ready="uploadReady"
+      :splitting="splitting"
+      :split-completed="splitCompleted"
+      :expand="false"
+      :disabled="!hasConfiguredModel"
+      @upload="(f) => emit('upload', f)"
+      @remove-file="(f) => emit('remove-file', f)"
+      class="w-full max-w-2xl"
+    />
+
+    <!-- 已上传文件 -->
+    <FileList
+      class="w-full max-w-2xl"
+      :pending-files="pendingFiles"
+      :file-progress="fileProgress"
+      :waiting-keys="waitingKeys"
+      :upload-busy="uploadBusy"
+      :upload-ready="uploadReady"
+      :splitting="splitting"
+      :split-completed="splitCompleted"
+      @remove-file="(f) => emit('remove-file', f)"
+    />
+
+    <!-- 操作按钮 -->
+    <ActionBar
+      class="mt-4"
+      :split-enabled="splitEnabled"
+      :export-enabled="false"
+      :splitting="splitting"
+      :split-completed="splitCompleted"
+      :upload-mode="uploadMode"
+      :erase-enabled="eraseEnabled"
+      @split="emit('split')"
+    />
+  </div>
+</template>
diff --git a/frontend/src/components/workspace/WorkspaceBackground.vue b/frontend/src/components/workspace/WorkspaceBackground.vue
new file mode 100644
index 00000000..8b8e9fca
--- /dev/null
+++ b/frontend/src/components/workspace/WorkspaceBackground.vue
@@ -0,0 +1,64 @@
+<script setup>
+/**
+ * WorkspaceBackground.vue
+ * 工作台全局背景装饰（Linear 风格光圈 + 噪点 + 闪烁星星）
+ */
+const bgStars = (() => {
+  const list = []
+  for (let i = 0; i < 50; i++) {
+    list.push({
+      left: Math.random() * 100, top: Math.random() * 100,
+      size: 0.5 + Math.random() * 2, opacity: 0.1 + Math.random() * 0.4,
+      duration: 2 + Math.random() * 4, delay: Math.random() * 5,
+    })
+  }
+  return list
+})()
+</script>
+
+<template>
+  <div class="fixed inset-0 z-0 pointer-events-none">
+    <div class="ws-bg-glow"></div>
+    <div class="ws-bg-noise"></div>
+    <div
+      v-for="(s, i) in bgStars"
+      :key="i"
+      class="absolute rounded-full bg-white ws-star"
+      :style="{
+        left: s.left + '%',
+        top: s.top + '%',
+        width: s.size + 'px',
+        height: s.size + 'px',
+        '--star-opacity': s.opacity,
+        animationDuration: s.duration + 's',
+        animationDelay: s.delay + 's',
+      }"
+    ></div>
+  </div>
+</template>
+
+<style>
+@keyframes ws-star-twinkle {
+  0%, 100% { opacity: var(--star-opacity); }
+  50% { opacity: 0.02; }
+}
+.ws-star {
+  animation: ws-star-twinkle ease-in-out infinite;
+}
+.ws-bg-glow {
+  position: absolute;
+  top: -20%;
+  left: -10%;
+  width: 600px;
+  height: 600px;
+  border-radius: 9999px;
+  background: radial-gradient(circle, rgba(129,115,223,0.06) 0%, transparent 70%);
+}
+.ws-bg-noise {
+  position: absolute;
+  inset: 0;
+  opacity: 0.04;
+  background-image: url("data:image/svg+xml,%3Csvg viewBox='0 0 256 256' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.85' numOctaves='4' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E");
+  background-size: 256px 256px;
+}
+</style>
diff --git a/frontend/src/composables/useAiChatSessions.js b/frontend/src/composables/useAiChatSessions.js
new file mode 100644
index 00000000..0eff78f6
--- /dev/null
+++ b/frontend/src/composables/useAiChatSessions.js
@@ -0,0 +1,89 @@
+import { ref, nextTick } from 'vue'
+import * as api from '@/api.js'
+
+export function useAiChatSessions(pushToast) {
+  const aiChatSessions = ref([])
+  const activeAiChatId = ref(null)
+  const chatMenuOpenId = ref(null)
+  const renamingChatId = ref(null)
+  const renameText = ref('')
+
+  async function loadAiChatSessions() {
+    try {
+      const data = await api.fetchMyChatSessions({ limit: 50 })
+      aiChatSessions.value = data.sessions || []
+    } catch (_) {}
+  }
+
+  async function createAiChat(currentView) {
+    try {
+      const session = await api.createIndependentChat('新对话')
+      aiChatSessions.value.unshift(session)
+      activeAiChatId.value = session.id
+      if (currentView.value !== 'ai-chat') currentView.value = 'ai-chat'
+    } catch (e) {
+      pushToast('error', e.message)
+    }
+  }
+
+  function selectAiChat(s, currentView) {
+    activeAiChatId.value = s.id
+    if (currentView.value !== 'ai-chat') currentView.value = 'ai-chat'
+  }
+
+  async function onAiChatTitleUpdated(sessionId, title) {
+    const s = aiChatSessions.value.find(s => s.id === sessionId)
+    if (s && s.title === '新对话') {
+      s.title = title
+      try { await api.updateChatTitle(sessionId, title) } catch (_) {}
+    }
+  }
+
+  function toggleChatMenu(id) {
+    chatMenuOpenId.value = chatMenuOpenId.value === id ? null : id
+  }
+
+  function closeChatMenu() {
+    chatMenuOpenId.value = null
+  }
+
+  async function startRenameChat(s) {
+    chatMenuOpenId.value = null
+    renamingChatId.value = s.id
+    renameText.value = s.title
+    await nextTick()
+    const input = document.querySelector('input[data-rename-input]')
+    if (input) { input.focus(); input.selectionStart = input.selectionEnd = input.value.length }
+  }
+
+  async function confirmRenameChat(s) {
+    const title = renameText.value.trim()
+    if (title && title !== s.title) {
+      try {
+        await api.updateChatTitle(s.id, title)
+        s.title = title
+      } catch (e) {
+        pushToast('error', e.message)
+      }
+    }
+    renamingChatId.value = null
+  }
+
+  async function deleteAiChat(id) {
+    try {
+      await api.deleteChat(id)
+      aiChatSessions.value = aiChatSessions.value.filter(s => s.id !== id)
+      if (activeAiChatId.value === id) activeAiChatId.value = null
+    } catch (e) {
+      pushToast('error', e.message)
+    }
+  }
+
+  return {
+    aiChatSessions, activeAiChatId,
+    chatMenuOpenId, renamingChatId, renameText,
+    loadAiChatSessions, createAiChat, selectAiChat,
+    onAiChatTitleUpdated, toggleChatMenu, closeChatMenu,
+    startRenameChat, confirmRenameChat, deleteAiChat,
+  }
+}
diff --git a/frontend/src/composables/useChatSession.js b/frontend/src/composables/useChatSession.js
new file mode 100644
index 00000000..15717017
--- /dev/null
+++ b/frontend/src/composables/useChatSession.js
@@ -0,0 +1,75 @@
+/**
+ * useChatSession.js
+ * AI 辅导对话会话管理（题目绑定对话）— 单例 composable
+ */
+import { ref } from 'vue'
+import * as api from '@/api.js'
+import { useToast } from '@/composables/useToast.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+
+// ── 模块级单例状态 ──────────────────────────────────────
+const chatSessionId = ref(null)
+const chatQuestion = ref(null)
+const chatActive = ref(false)
+const answerModalOpen = ref(false)
+const answerModalTarget = ref(null)
+const answerModalText = ref('')
+const answerModalSaving = ref(false)
+
+export function useChatSession() {
+  const { pushToast } = useToast()
+  const { currentView } = useWorkspaceNav()
+
+  const doOpenChatSession = async (question) => {
+    try {
+      const sessions = await api.fetchChatSessions(question.id)
+      chatSessionId.value = sessions.length ? sessions[0].id : (await api.createChat(question.id)).id
+      chatActive.value = true
+      currentView.value = 'chat'
+    } catch (e) {
+      pushToast('error', '打开对话失败: ' + (e instanceof Error ? e.message : String(e)))
+    }
+  }
+
+  const openChat = async (question) => {
+    chatQuestion.value = question
+    if (!question.answer) {
+      answerModalTarget.value = question
+      answerModalText.value = ''
+      answerModalOpen.value = true
+      return
+    }
+    await doOpenChatSession(question)
+  }
+
+  const saveAnswerAndChat = async () => {
+    if (!answerModalTarget.value || answerModalSaving.value) return
+    const text = answerModalText.value.trim()
+    if (!text) { pushToast('error', '请输入答案/解析内容'); return }
+    answerModalSaving.value = true
+    try {
+      await api.saveQuestionAnswer(answerModalTarget.value.id, text)
+      answerModalTarget.value.answer = text
+      answerModalOpen.value = false
+      pushToast('success', '答案已保存')
+      await doOpenChatSession(answerModalTarget.value)
+    } catch (e) {
+      pushToast('error', '保存答案失败: ' + (e instanceof Error ? e.message : String(e)))
+    } finally {
+      answerModalSaving.value = false
+    }
+  }
+
+  const backToErrorBank = () => {
+    chatActive.value = false
+    chatSessionId.value = null
+    chatQuestion.value = null
+    currentView.value = 'error-bank'
+  }
+
+  return {
+    chatSessionId, chatQuestion, chatActive,
+    answerModalOpen, answerModalTarget, answerModalText, answerModalSaving,
+    openChat, saveAnswerAndChat, backToErrorBank,
+  }
+}
diff --git a/frontend/src/composables/useFileUpload.js b/frontend/src/composables/useFileUpload.js
new file mode 100644
index 00000000..9c16885c
--- /dev/null
+++ b/frontend/src/composables/useFileUpload.js
@@ -0,0 +1,175 @@
+import { ref, reactive, computed } from 'vue'
+import { fileKey } from '@/utils.js'
+import * as api from '@/api.js'
+
+export function useFileUpload(pushToast, S, questions, selectedIds, splitCompleted, splitting, uploadMode) {
+  const uploadBusy = ref(false)
+  const uploadReady = ref(false)
+  const pendingFiles = reactive([])
+  const pendingPreviewUrls = computed(() =>
+    pendingFiles.filter(pf => pf.file).map(pf => URL.createObjectURL(pf.file))
+  )
+  const fileProgress = reactive({})
+  const waitingKeys = reactive(new Set())
+  const uploadQueue = reactive([])
+  let activeXhr = null
+  let fakeProgressTimer = null
+  let fakeProgressKeys = []
+
+  const setProgress = (key, p) => { fileProgress[key] = Math.max(0, Math.min(100, Number(p) || 0)) }
+
+  const stopFakeProgress = () => {
+    if (fakeProgressTimer) {
+      window.clearInterval(fakeProgressTimer)
+      fakeProgressTimer = null
+    }
+    fakeProgressKeys = []
+  }
+
+  const startFakeProgress = (keys) => {
+    stopFakeProgress()
+    fakeProgressKeys = Array.from(keys || [])
+    const tick = () => {
+      if (!uploadBusy.value) { stopFakeProgress(); return }
+      for (const key of fakeProgressKeys) {
+        const current = Number(fileProgress[key] || 0)
+        const cap = 82
+        if (current >= cap) continue
+        let inc = current < 55 ? 1 + Math.random() * 3 : current < 75 ? 0.4 + Math.random() * 1.1 : 0.08 + Math.random() * 0.25
+        setProgress(key, Math.min(cap, current + inc))
+      }
+    }
+    tick()
+    fakeProgressTimer = window.setInterval(tick, 360)
+  }
+
+  const handleUpload = (files) => {
+    const uploadFiles = Array.from(files || []).filter(f => pendingFiles.some(x => x.key === fileKey(f)))
+    if (!uploadFiles.length) return
+    uploadBusy.value = true
+    const step = S.value
+    const keys = uploadFiles.map(f => fileKey(f))
+    for (const k of keys) waitingKeys.delete(k)
+    startFakeProgress(keys)
+
+    const formData = new FormData()
+    for (const f of uploadFiles) {
+      formData.append('files', f)
+      formData.append('file_key', fileKey(f))
+    }
+
+    activeXhr = api.uploadFiles(formData, {
+      onProgress: (ratio) => {
+        const pct = Math.max(0, Math.min(95, ratio * 95))
+        for (const k of keys) {
+          if (pendingFiles.some(x => x.key === k)) setProgress(k, Math.max(fileProgress[k] || 0, pct))
+        }
+      },
+      onSuccess: () => {
+        stopFakeProgress()
+        uploadBusy.value = false
+        activeXhr = null
+        for (const k of keys) {
+          if (pendingFiles.some(x => x.key === k)) setProgress(k, 100)
+        }
+        uploadReady.value = pendingFiles.length > 0
+        pushToast('success', '上传成功')
+        pumpUploadQueue()
+      },
+      onError: (msg) => {
+        stopFakeProgress()
+        uploadBusy.value = false
+        activeXhr = null
+        pushToast('error', msg)
+        pumpUploadQueue()
+      },
+      onAbort: () => {
+        stopFakeProgress()
+        uploadBusy.value = false
+        activeXhr = null
+        pumpUploadQueue()
+      },
+    })
+  }
+
+  const pumpUploadQueue = () => {
+    if (uploadBusy.value || !uploadQueue.length) return
+    const next = uploadQueue.shift()
+    if (next && next.length) handleUpload(next)
+  }
+
+  const enqueueUpload = (files) => {
+    const list = Array.from(files || [])
+    if (!list.length) return
+    if (splitCompleted.value || splitting.value) { pushToast('error', '已分割完成，请先重新开始'); return }
+    for (const f of list) {
+      const k = fileKey(f)
+      if (pendingFiles.some(x => x.key === k)) continue
+      pendingFiles.push({ key: k, file: f })
+      setProgress(k, 0)
+    }
+    if (uploadBusy.value) {
+      for (const f of list) waitingKeys.add(fileKey(f))
+      uploadQueue.push(list)
+      return
+    }
+    handleUpload(list)
+  }
+
+  const doCancelFile = async (key, step) => {
+    try {
+      if (fakeProgressKeys.length) fakeProgressKeys = fakeProgressKeys.filter(k => k !== key)
+      if (!fakeProgressKeys.length) stopFakeProgress()
+      const data = await api.cancelFile(key)
+      const idx = pendingFiles.findIndex(x => x.key === key)
+      if (idx >= 0) pendingFiles.splice(idx, 1)
+      delete fileProgress[key]
+      waitingKeys.delete(key)
+      questions.value = []
+      selectedIds.clear()
+      splitCompleted.value = false
+      if (!pendingFiles.length) {
+        uploadReady.value = false
+      }
+      pushToast('success', data.message || '已撤销')
+      if (!pendingFiles.length && activeXhr) {
+        try { activeXhr.abort() } catch (_) {}
+      }
+    } catch (_) { pushToast('error', '撤销失败: 网络错误') }
+  }
+
+  const removePendingFile = async (key) => {
+    if (!key || splitting.value || splitCompleted.value) return
+    if (uploadBusy.value || uploadReady.value) { await doCancelFile(key, S.value); return }
+    const idx = pendingFiles.findIndex(x => x.key === key)
+    if (idx >= 0) pendingFiles.splice(idx, 1)
+    delete fileProgress[key]
+    waitingKeys.delete(key)
+    for (let i = uploadQueue.length - 1; i >= 0; i--) {
+      uploadQueue[i] = (uploadQueue[i] || []).filter(f => fileKey(f) !== key)
+      if (!uploadQueue[i].length) uploadQueue.splice(i, 1)
+    }
+  }
+
+  const doReset = (providerOptions, selectedModel, step) => {
+    uploadBusy.value = false
+    uploadReady.value = false
+    splitting.value = false
+    splitCompleted.value = false
+    pendingFiles.splice(0, pendingFiles.length)
+    for (const k of Object.keys(fileProgress)) delete fileProgress[k]
+    waitingKeys.clear()
+    uploadQueue.splice(0, uploadQueue.length)
+    questions.value = []
+    selectedIds.clear()
+    const configured = providerOptions.value.find(m => m.configured)
+    selectedModel.value = configured ? configured.default_model : ''
+    pushToast('success', '已重置')
+  }
+
+  return {
+    uploadBusy, uploadReady, pendingFiles, pendingPreviewUrls,
+    fileProgress, waitingKeys,
+    enqueueUpload, removePendingFile, stopFakeProgress, doReset,
+  }
+}
diff --git a/frontend/src/composables/useImageModal.js b/frontend/src/composables/useImageModal.js
new file mode 100644
index 00000000..66a5797b
--- /dev/null
+++ b/frontend/src/composables/useImageModal.js
@@ -0,0 +1,28 @@
+import { ref } from 'vue'
+
+// ── 模块级单例状态 ──────────────────────────────────────
+const modalOpen = ref(false)
+const modalSrc = ref('')
+const modalScale = ref(1)
+
+const openModal = (src) => {
+  modalSrc.value = src || ''
+  modalScale.value = 1
+  modalOpen.value = !!src
+  if (src) document.body.style.overflow = 'hidden'
+}
+
+const closeModal = () => {
+  modalOpen.value = false
+  modalSrc.value = ''
+  modalScale.value = 1
+  document.body.style.overflow = ''
+}
+
+/**
+ * useImageModal — 图片预览弹窗单例 composable
+ * 任何组件调用 useImageModal() 都操作同一个弹窗实例
+ */
+export function useImageModal() {
+  return { modalOpen, modalSrc, modalScale, openModal, closeModal }
+}
diff --git a/frontend/src/composables/usePageTransition.js b/frontend/src/composables/usePageTransition.js
index bb83c026..8c2d7792 100644
--- a/frontend/src/composables/usePageTransition.js
+++ b/frontend/src/composables/usePageTransition.js
@@ -11,7 +11,7 @@ let resolveEnter = null
 export function usePageTransition() {
   /**
    * 显示遮罩
-   * 返回一个 Promise，该 Promise 将在 AppLoading 组件触发 @after-enter 后 resolve
+   * 返回一个 Promise，该 Promise 将在 BaseLoading 组件触发 @after-enter 后 resolve
    */
   const show = () => {
     if (hideTimer) {
@@ -38,7 +38,7 @@ export function usePageTransition() {
   }
 
   /**
-   * 遮罩淡入动画完成后的回调，由 App.vue 监听 AppLoading 事件后调用
+   * 遮罩淡入动画完成后的回调，由 App.vue 监听 BaseLoading 事件后调用
    */
   const notifyEnterCompleted = () => {
     if (resolveEnter) {
diff --git a/frontend/src/composables/usePaginatedList.js b/frontend/src/composables/usePaginatedList.js
new file mode 100644
index 00000000..1bc8389e
--- /dev/null
+++ b/frontend/src/composables/usePaginatedList.js
@@ -0,0 +1,81 @@
+import { ref, computed, watch, onBeforeUnmount } from 'vue'
+
+/**
+ * usePaginatedList — 分页 + 筛选 + 防抖查询 composable
+ * @param {Function} fetchFn - 查询函数 (params) => { items, total, total_pages? }
+ * @param {Object} options
+ * @param {number} options.pageSize - 每页条数，默认 10
+ * @param {number} options.filterDebounce - 筛选防抖延迟（ms），默认 300
+ * @param {number} options.keywordDebounce - 关键字防抖延迟（ms），默认 500
+ */
+export function usePaginatedList(fetchFn, { pageSize: defaultPageSize = 10, filterDebounce = 300, keywordDebounce = 500 } = {}) {
+  const items = ref([])
+  const page = ref(1)
+  const pageSize = ref(defaultPageSize)
+  const total = ref(0)
+  const totalPages = computed(() => Math.ceil(total.value / pageSize.value))
+  const loading = ref(false)
+
+  let debounceTimer = null
+
+  const doQuery = async (params = {}) => {
+    loading.value = true
+    try {
+      const data = await fetchFn({ page: page.value, page_size: pageSize.value, ...params })
+      items.value = data.items || []
+      total.value = data.total || 0
+      return data
+    } finally {
+      loading.value = false
+    }
+  }
+
+  const goPage = (p) => {
+    if (p < 1 || p > totalPages.value) return
+    page.value = p
+  }
+
+  /**
+   * 监听筛选变化，自动重置页码并防抖查询
+   * @param {Function} source - watch source（返回筛选依赖数组）
+   * @param {Function} buildParams - 构建查询参数
+   * @param {Object} opts
+   * @param {boolean} opts.isKeyword - 是否使用关键字防抖延迟
+   * @param {Function} opts.afterQuery - 查询完成后的回调
+   */
+  const watchFilters = (source, buildParams, { isKeyword = false, afterQuery } = {}) => {
+    const delay = isKeyword ? keywordDebounce : filterDebounce
+    watch(source, () => {
+      clearTimeout(debounceTimer)
+      debounceTimer = setTimeout(async () => {
+        page.value = 1
+        const data = await doQuery(buildParams())
+        afterQuery?.(data)
+      }, delay)
+    })
+  }
+
+  const watchPage = (buildParams, { afterQuery } = {}) => {
+    watch(page, async () => {
+      const data = await doQuery(buildParams())
+      afterQuery?.(data)
+    })
+  }
+
+  onBeforeUnmount(() => {
+    if (debounceTimer) clearTimeout(debounceTimer)
+  })
+
+  return {
+    items,
+    page,
+    pageSize,
+    total,
+    totalPages,
+    loading,
+    doQuery,
+    goPage,
+    watchFilters,
+    watchPage,
+  }
+}
diff --git a/frontend/src/composables/useQuestionList.js b/frontend/src/composables/useQuestionList.js
new file mode 100644
index 00000000..2741ef72
--- /dev/null
+++ b/frontend/src/composables/useQuestionList.js
@@ -0,0 +1,30 @@
+import { ref, reactive, nextTick } from 'vue'
+import { typesetMath as _typesetMathEl } from '@/utils.js'
+
+export function useQuestionList() {
+  const questions = ref([])
+  const selectedIds = reactive(new Set())
+  const questionListRef = ref(null)
+
+  const toggleQuestion = (id) => { selectedIds.has(id) ? selectedIds.delete(id) : selectedIds.add(id) }
+  const selectAll = () => { for (const q of questions.value) selectedIds.add(q.uid) }
+  const deselectAll = () => { selectedIds.clear() }
+
+  const reorderQuestions = (oldIndex, newIndex) => {
+    const arr = questions.value.slice()
+    const [moved] = arr.splice(oldIndex, 1)
+    arr.splice(newIndex, 0, moved)
+    questions.value = arr
+  }
+
+  const typesetMath = async () => {
+    await nextTick()
+    const el = questionListRef.value?.questionsBoxEl
+    await _typesetMathEl(el || undefined)
+  }
+
+  return {
+    questions, selectedIds, questionListRef,
+    toggleQuestion, selectAll, deselectAll, reorderQuestions, typesetMath,
+  }
+}
diff --git a/frontend/src/composables/useSelectableList.js b/frontend/src/composables/useSelectableList.js
new file mode 100644
index 00000000..97e40807
--- /dev/null
+++ b/frontend/src/composables/useSelectableList.js
@@ -0,0 +1,38 @@
+import { reactive, ref } from 'vue'
+
+/**
+ * useSelectableList — 列表多选模式 composable
+ * @param {Object} options
+ * @param {string} options.idKey - 项目 ID 字段名，默认 'id'
+ * @returns {{ selectMode, selectedIds, toggleSelectMode, toggleSelect, selectAllItems, clearSelection }}
+ */
+export function useSelectableList({ idKey = 'id' } = {}) {
+  const selectMode = ref(false)
+  const selectedIds = reactive(new Set())
+
+  const toggleSelectMode = () => {
+    selectMode.value = !selectMode.value
+    if (!selectMode.value) selectedIds.clear()
+  }
+
+  const toggleSelect = (id) => {
+    selectedIds.has(id) ? selectedIds.delete(id) : selectedIds.add(id)
+  }
+
+  const selectAllItems = (items) => {
+    items.forEach((item) => selectedIds.add(item[idKey]))
+  }
+
+  const clearSelection = () => {
+    selectedIds.clear()
+  }
+
+  return {
+    selectMode,
+    selectedIds,
+    toggleSelectMode,
+    toggleSelect,
+    selectAllItems,
+    clearSelection,
+  }
+}
diff --git a/frontend/src/composables/useSidebarIndicator.js b/frontend/src/composables/useSidebarIndicator.js
new file mode 100644
index 00000000..93575de0
--- /dev/null
+++ b/frontend/src/composables/useSidebarIndicator.js
@@ -0,0 +1,76 @@
+import { ref } from 'vue'
+
+export function useSidebarIndicator() {
+  const navRef = ref(null)
+  const navBtnRefs = ref({})
+  const indicatorStyle = ref({ opacity: 0, top: '0px', height: '0px' })
+  const indicatorTransition = ref(true)
+
+  const chatListRef = ref(null)
+  const chatBtnRefs = ref({})
+  const chatIndicatorStyle = ref({ opacity: 0, top: '0px', height: '0px' })
+  const chatIndicatorTransition = ref(true)
+
+  function updateIndicator(currentView, activeAiChatId, NAV_GROUPS, collapsedGroups, animate = true) {
+    const cv = currentView
+    const isChat = cv === 'ai-chat'
+
+    // === 导航组指示器 ===
+    if (isChat) {
+      indicatorTransition.value = animate
+      indicatorStyle.value = { opacity: 0, top: '0px', height: '0px' }
+    } else {
+      let matchId = null
+      let matchGroupIdx = -1
+      for (let gi = 0; gi < NAV_GROUPS.length; gi++) {
+        for (const item of NAV_GROUPS[gi].items) {
+          if (item.match && item.match(cv)) { matchId = item.id; matchGroupIdx = gi; break }
+        }
+        if (matchId) break
+      }
+      if (matchGroupIdx >= 0 && collapsedGroups[matchGroupIdx]) {
+        indicatorTransition.value = false
+        indicatorStyle.value = { opacity: 0, top: '0px', height: '0px' }
+      } else if (!matchId || !navBtnRefs.value[matchId] || !navRef.value) {
+        indicatorTransition.value = animate
+        indicatorStyle.value = { opacity: 0, top: '0px', height: '0px' }
+      } else {
+        indicatorTransition.value = animate
+        const navRect = navRef.value.getBoundingClientRect()
+        const btnRect = navBtnRefs.value[matchId].getBoundingClientRect()
+        indicatorStyle.value = {
+          opacity: 1,
+          top: (btnRect.top - navRect.top) + 'px',
+          height: btnRect.height + 'px',
+        }
+      }
+    }
+
+    // === 对话区指示器 ===
+    if (!isChat || !activeAiChatId) {
+      chatIndicatorTransition.value = animate
+      chatIndicatorStyle.value = { opacity: 0, top: '0px', height: '0px' }
+    } else {
+      const btnEl = chatBtnRefs.value[activeAiChatId]
+      if (!btnEl || !chatListRef.value) {
+        chatIndicatorTransition.value = animate
+        chatIndicatorStyle.value = { opacity: 0, top: '0px', height: '0px' }
+      } else {
+        chatIndicatorTransition.value = animate
+        const listRect = chatListRef.value.getBoundingClientRect()
+        const btnRect = btnEl.getBoundingClientRect()
+        chatIndicatorStyle.value = {
+          opacity: 1,
+          top: (btnRect.top - listRect.top) + 'px',
+          height: btnRect.height + 'px',
+        }
+      }
+    }
+  }
+
+  return {
+    navRef, navBtnRefs, indicatorStyle, indicatorTransition,
+    chatListRef, chatBtnRefs, chatIndicatorStyle, chatIndicatorTransition,
+    updateIndicator,
+  }
+}
diff --git a/frontend/src/composables/useSplitPipeline.js b/frontend/src/composables/useSplitPipeline.js
new file mode 100644
index 00000000..0330ab9a
--- /dev/null
+++ b/frontend/src/composables/useSplitPipeline.js
@@ -0,0 +1,179 @@
+import { ref } from 'vue'
+import * as api from '@/api.js'
+
+export function useSplitPipeline(pushToast, currentView, step, S, uploadReady, splitting, splitCompleted, uploadMode, selectedProvider, selectedModel, questions, selectedIds, pendingFiles, typesetMath) {
+  const eraseLoading = ref(false)
+  const eraseImages = ref([])
+  const eraseDone = ref(false)
+  const ocrLoading = ref(false)
+  const ocrPages = ref([])
+  const ocrDone = ref(false)
+  const eraseEnabled = ref(true)
+
+  const startProcess = () => {
+    if (eraseEnabled.value) {
+      doErase()
+    } else {
+      doOcr()
+    }
+  }
+
+  const doErase = async () => {
+    if (eraseLoading.value) return
+    eraseLoading.value = true
+    eraseDone.value = false
+    eraseImages.value = []
+    currentView.value = 'workspace_review'
+    step.value = S.value.ERASE
+    try {
+      const data = await api.runErase()
+      eraseImages.value = data.images || []
+      eraseDone.value = true
+      pushToast('success', data.message || '擦除完成')
+    } catch (e) {
+      pushToast('error', e.message || '擦除失败')
+      currentView.value = 'workspace'
+      step.value = S.value.UPLOAD
+    } finally {
+      eraseLoading.value = false
+    }
+  }
+
+  const doOcr = async () => {
+    if (ocrLoading.value) return
+    ocrLoading.value = true
+    ocrDone.value = false
+    ocrPages.value = []
+    eraseDone.value = false
+    currentView.value = 'workspace_review'
+    step.value = S.value.OCR
+    try {
+      const data = await api.runOcr()
+      ocrPages.value = data.pages || []
+      ocrDone.value = true
+      pushToast('success', `OCR 完成，共 ${data.pages?.length || 0} 页`)
+    } catch (e) {
+      pushToast('error', e.message || 'OCR 失败')
+      currentView.value = 'workspace'
+      step.value = S.value.UPLOAD
+    } finally {
+      ocrLoading.value = false
+    }
+  }
+
+  const doSplit = async () => {
+    if (!uploadReady.value || splitting.value || splitCompleted.value) return
+
+    if (uploadMode.value === 'note') {
+      await doNoteOrganize()
+      return
+    }
+
+    splitting.value = true
+    ocrDone.value = false
+    step.value = S.value.SPLIT
+    pushToast('info', '正在调用AI分割题目，请稍候...', 1800)
+    try {
+      const data = await api.splitQuestions(selectedProvider.value, selectedModel.value)
+      questions.value = data.questions || []
+      selectedIds.clear()
+      if (data.warnings && data.warnings.length) {
+        for (const w of data.warnings) pushToast('warning', w, 6000)
+      }
+      if (questions.value.length > 0) {
+        splitCompleted.value = true
+        step.value = S.value.EXPORT
+        if (!data.warnings || !data.warnings.length) {
+          pushToast('success', `成功分割 ${questions.value.length} 道题目`)
+        }
+        await typesetMath()
+        setTimeout(() => {
+          if (currentView.value === 'workspace') {
+            currentView.value = 'workspace_review'
+          }
+        }, 800)
+      }
+    } catch (e) {
+      pushToast('error', '分割失败: ' + (e instanceof Error ? e.message : String(e)))
+    } finally {
+      splitting.value = false
+    }
+  }
+
+  const doNoteOrganize = async () => {
+    splitting.value = true
+    step.value = S.value.SPLIT
+    pushToast('info', '正在调用AI整理笔记，请稍候...', 3000)
+    try {
+      const formData = new FormData()
+      for (const pf of pendingFiles) {
+        if (pf.file) formData.append('files', pf.file)
+      }
+      formData.append('model_provider', selectedProvider.value)
+      if (selectedModel.value) formData.append('model_name', selectedModel.value)
+
+      const result = await new Promise((resolve, reject) => {
+        api.createNote(formData, {
+          onSuccess: (data) => resolve(data),
+          onError: (msg) => reject(new Error(msg)),
+        })
+      })
+
+      splitCompleted.value = true
+      step.value = S.value.EXPORT
+      pushToast('success', '笔记整理完成！')
+
+      setTimeout(() => {
+        currentView.value = 'notes'
+      }, 800)
+    } catch (e) {
+      pushToast('error', '笔记整理失败: ' + (e instanceof Error ? e.message : String(e)))
+    } finally {
+      splitting.value = false
+    }
+  }
+
+  const doExport = async () => {
+    if (!selectedIds.size) { pushToast('error', '请至少选择一道题目！'); return }
+    try {
+      const data = await api.exportQuestions(Array.from(selectedIds))
+      step.value = S.value.EXPORT + 1
+      pushToast('success', `错题本导出成功！已保存到: ${data.output_path}`)
+      let filename = 'wrongbook.md'
+      if (data.output_path) {
+        const parts = String(data.output_path).split(/[/\\]/)
+        const last = parts[parts.length - 1]
+        if (last) filename = last
+      }
+      const a = document.createElement('a')
+      a.href = `/download/${encodeURIComponent(filename)}?t=${Date.now()}`
+      a.download = filename
+      a.style.display = 'none'
+      document.body.appendChild(a)
+      a.click()
+      a.remove()
+    } catch (e) {
+      pushToast('error', '导出失败: ' + (e instanceof Error ? e.message : String(e)))
+    }
+  }
+
+  const doSaveToDb = async (errorBankRef) => {
+    if (!selectedIds.size) { pushToast('error', '请至少选择一道题目！'); return }
+    try {
+      const answers = questions.value
+        .filter(q => selectedIds.has(q.uid) && (q.answer || q.user_answer))
+        .map(q => ({ uid: q.uid, answer: q.answer || '', user_answer: q.user_answer || '' }))
+      const data = await api.saveToDb(Array.from(selectedIds), answers)
+      pushToast('success', data.message || '已导入错题库')
+      errorBankRef?.value?.refresh()
+    } catch (e) {
+      pushToast('error', '导入失败: ' + (e instanceof Error ? e.message : String(e)))
+    }
+  }
+
+  return {
+    eraseEnabled, eraseLoading, eraseImages, eraseDone,
+    ocrLoading, ocrPages, ocrDone,
+    startProcess, doErase, doOcr, doSplit, doExport, doSaveToDb,
+  }
+}
diff --git a/frontend/src/composables/useSystemStatus.js b/frontend/src/composables/useSystemStatus.js
new file mode 100644
index 00000000..a913a1f9
--- /dev/null
+++ b/frontend/src/composables/useSystemStatus.js
@@ -0,0 +1,74 @@
+import { ref, computed, watch } from 'vue'
+import * as api from '@/api.js'
+
+// ── 模块级单例状态 ──────────────────────────────────────
+const statusLoading = ref(true)
+const systemStatus = ref(null)
+const statusError = ref('')
+const selectedModel = ref('')
+
+const providerOptions = computed(() => {
+  const s = systemStatus.value
+  return s && s.available_models ? s.available_models : []
+})
+
+const hasConfiguredModel = computed(() => providerOptions.value.some(p => p.configured))
+
+const selectedProvider = computed(() => {
+  for (const p of providerOptions.value) {
+    if (p.models && p.models.includes(selectedModel.value)) return p.value
+  }
+  const configured = providerOptions.value.find(p => p.configured)
+  return configured ? configured.value : (providerOptions.value[0]?.value ?? 'openai')
+})
+
+watch(systemStatus, (newVal) => {
+  if (newVal && newVal.available_models) {
+    const configured = newVal.available_models.find(m => m.configured)
+    if (configured) selectedModel.value = configured.default_model
+  }
+})
+
+const statusPills = computed(() => {
+  if (statusLoading.value) return [
+    { key: 'paddle', loading: true, label: 'PaddleOCR' },
+    { key: 'ensexam', loading: true, label: 'EnsExam' },
+    { key: 'langsmith', loading: true, label: 'LangSmith' },
+  ]
+  const s = systemStatus.value
+  if (!s) return []
+  const pills = []
+  pills.push({ key: 'paddle', ok: !!s.paddleocr_configured, label: s.paddleocr_configured ? 'PaddleOCR' : 'PaddleOCR未配置' })
+  if (s.ensexam_configured) {
+    pills.push({ key: 'ensexam', ok: true, label: 'EnsExam' })
+  }
+  pills.push(s.langsmith_enabled
+    ? { key: 'langsmith', ok: true, label: 'LangSmith追踪' }
+    : { key: 'langsmith', ok: false, label: 'LangSmith', isPlaceholder: true }
+  )
+  return pills
+})
+
+const doFetchStatus = async () => {
+  statusLoading.value = true
+  statusError.value = ''
+  try {
+    systemStatus.value = await api.fetchStatus()
+  } catch (e) {
+    statusError.value = e instanceof Error ? e.message : String(e)
+  } finally {
+    statusLoading.value = false
+  }
+}
+
+/**
+ * useSystemStatus — 系统状态单例 composable
+ * 任何组件调用都返回同一份响应式状态
+ */
+export function useSystemStatus() {
+  return {
+    statusLoading, systemStatus, statusError, selectedModel,
+    providerOptions, hasConfiguredModel, selectedProvider, statusPills,
+    doFetchStatus,
+  }
+}
diff --git a/frontend/src/composables/useTheme.js b/frontend/src/composables/useTheme.js
new file mode 100644
index 00000000..920bc79f
--- /dev/null
+++ b/frontend/src/composables/useTheme.js
@@ -0,0 +1,57 @@
+import { ref } from 'vue'
+
+const isDark = ref(document.documentElement.classList.contains('dark'))
+
+export function useTheme() {
+  function setTheme(dark) {
+    isDark.value = dark
+    document.documentElement.classList.toggle('dark', dark)
+    localStorage.setItem('theme', dark ? 'dark' : 'light')
+  }
+
+  /**
+   * 带圆形扩散动画的主题切换
+   * @param {HTMLElement} [btnEl] - 触发按钮元素，动画从此处扩散；不传则无动画
+   */
+  async function toggleTheme(btnEl) {
+    const nextDark = !isDark.value
+
+    const prefersReduce = window.matchMedia?.('(prefers-reduced-motion: reduce)').matches
+    const canTransition = !prefersReduce && typeof document.startViewTransition === 'function'
+
+    if (!canTransition || !btnEl) {
+      setTheme(nextDark)
+      return
+    }
+
+    const rect = btnEl.getBoundingClientRect()
+    const x = rect.left + rect.width / 2
+    const y = rect.top + rect.height / 2
+    const endRadius = Math.hypot(Math.max(x, window.innerWidth - x), Math.max(y, window.innerHeight - y))
+
+    const transition = document.startViewTransition(() => setTheme(nextDark))
+    try {
+      await transition.ready
+      const duration = 1000
+      document.documentElement.animate(
+        { clipPath: [`circle(0px at ${x}px ${y}px)`, `circle(${endRadius}px at ${x}px ${y}px)`] },
+        { duration, easing: 'cubic-bezier(0.2, 0, 0, 1)', pseudoElement: '::view-transition-new(root)' },
+      )
+      document.documentElement.animate(
+        { opacity: [1, 0.98] },
+        { duration, easing: 'linear', pseudoElement: '::view-transition-old(root)' },
+      )
+      await transition.finished
+    } catch (_) {
+      setTheme(nextDark)
+    }
+  }
+
+  function initTheme() {
+    const saved = localStorage.getItem('theme') || 'dark'
+    isDark.value = saved === 'dark'
+    document.documentElement.classList.toggle('dark', isDark.value)
+  }
+
+  return { isDark, setTheme, toggleTheme, initTheme }
+}
diff --git a/frontend/src/composables/useToast.js b/frontend/src/composables/useToast.js
new file mode 100644
index 00000000..756dacd4
--- /dev/null
+++ b/frontend/src/composables/useToast.js
@@ -0,0 +1,16 @@
+import { inject } from 'vue'
+
+const TOAST_KEY = 'pushToast'
+const noopToast = (type, msg) => { console.warn(`[Toast:${type}] ${msg}`) }
+
+/**
+ * useToast — 注入 pushToast，带安全 fallback
+ * 用于子组件中，替代手写 inject('pushToast', fallback)
+ */
+export function useToast() {
+  const pushToast = inject(TOAST_KEY, noopToast)
+  return { pushToast }
+}
+
+/** provide 时使用的 key，保持一致 */
+export const TOAST_INJECTION_KEY = TOAST_KEY
diff --git a/frontend/src/composables/useWorkspaceNav.js b/frontend/src/composables/useWorkspaceNav.js
new file mode 100644
index 00000000..b447651d
--- /dev/null
+++ b/frontend/src/composables/useWorkspaceNav.js
@@ -0,0 +1,90 @@
+/**
+ * useWorkspaceNav.js
+ * 工作台视图路由 + 侧边栏导航配置 — 单例 composable
+ */
+import { ref, computed, watch } from 'vue'
+import { useRouter, useRoute } from 'vue-router'
+
+const VIEW_TO_PATH = {
+  workspace: '/app/workspace',
+  workspace_review: '/app/workspace/review',
+  dashboard: '/app/dashboard',
+  review: '/app/review',
+  'error-bank': '/app/error-bank',
+  notes: '/app/notes',
+  'ai-chat': '/app/ai-chat',
+  settings: '/app/settings',
+  'split-history': '/app/split-history',
+  chat: '/app/chat',
+}
+
+const WORKSPACE_VIEWS = new Set(['workspace', 'workspace_review', 'split-history'])
+
+const NAV_GROUPS = [
+  {
+    label: null,
+    items: [
+      { id: 'workspace', label: '录入工作台', icon: 'fa-wand-magic-sparkles', match: (v) => WORKSPACE_VIEWS.has(v) },
+    ],
+  },
+  {
+    label: '数据',
+    collapsible: true,
+    items: [
+      { id: 'dashboard', label: '数据面板', icon: 'fa-chart-pie', match: (v) => v === 'dashboard' },
+      { id: 'error-bank', label: '错题库', icon: 'fa-database', match: (v) => v === 'error-bank' },
+      { id: 'notes', label: '笔记库', icon: 'fa-book-open', match: (v) => v === 'notes' },
+    ],
+  },
+  {
+    label: '更多',
+    collapsible: true,
+    items: [
+      { id: 'review-disabled', label: '刷题', icon: 'fa-clock-rotate-left', disabled: true },
+    ],
+  },
+]
+
+// ── 模块级单例状态 ──────────────────────────────────────
+const collapsedGroups = ref({})
+const chatCollapsed = ref(false)
+const lastWorkspaceView = ref('workspace')
+let initialized = false
+
+export function useWorkspaceNav() {
+  const router = useRouter()
+  const route = useRoute()
+
+  const currentView = computed({
+    get() {
+      const view = route.params.view || 'workspace'
+      const subview = route.params.subview
+      if (view === 'workspace' && subview === 'review') return 'workspace_review'
+      return view
+    },
+    set(view) {
+      const path = VIEW_TO_PATH[view]
+      if (path && route.path !== path) router.push(path)
+    },
+  })
+
+  if (!initialized) {
+    initialized = true
+    watch(currentView, (v) => {
+      if (WORKSPACE_VIEWS.has(v)) lastWorkspaceView.value = v
+    })
+  }
+
+  const navigateToHome = () => {
+    document.body.style.transition = 'opacity 0.25s ease, transform 0.25s ease'
+    document.body.style.opacity = '0'
+    document.body.style.transform = 'translateY(-6px)'
+    setTimeout(() => { window.location.href = '/' }, 260)
+  }
+
+  return {
+    currentView, lastWorkspaceView, collapsedGroups, chatCollapsed,
+    NAV_GROUPS, WORKSPACE_VIEWS,
+    navigateToHome,
+  }
+}
diff --git a/frontend/src/composables/useWorkspaceToast.js b/frontend/src/composables/useWorkspaceToast.js
new file mode 100644
index 00000000..d94504a2
--- /dev/null
+++ b/frontend/src/composables/useWorkspaceToast.js
@@ -0,0 +1,18 @@
+import { ref } from 'vue'
+
+export function useWorkspaceToast() {
+  const toasts = ref([])
+  let toastId = 0
+
+  const pushToast = (type, message, timeout = 2600) => {
+    const id = ++toastId
+    toasts.value = [{ id, type, message }, ...toasts.value].slice(0, 5)
+    if (timeout > 0) {
+      window.setTimeout(() => {
+        toasts.value = toasts.value.filter(t => t.id !== id)
+      }, timeout)
+    }
+  }
+
+  return { toasts, pushToast }
+}
diff --git a/frontend/src/main.js b/frontend/src/main.js
index 4806c143..770fd34a 100644
--- a/frontend/src/main.js
+++ b/frontend/src/main.js
@@ -1,6 +1,6 @@
 import { createApp } from 'vue'
-import './style.css'
-import App from './App.vue'
-import router from './router/index.js'
+import '@/style.css'
+import App from '@/App.vue'
+import router from '@/router/index.js'
 
 createApp(App).use(router).mount('#app')
diff --git a/frontend/src/router/index.js b/frontend/src/router/index.js
index d7c59716..62cd4bf6 100644
--- a/frontend/src/router/index.js
+++ b/frontend/src/router/index.js
@@ -1,29 +1,41 @@
+/**
+ * router/index.js
+ * Vue Router 路由配置
+ *
+ * 三种布局：home（首页）、auth（认证页）、app（工作台）
+ * 路由守卫：认证检查 + 跨布局过渡动画
+ */
 import { nextTick } from 'vue'
 import { createRouter, createWebHistory } from 'vue-router'
-import { useAuth } from '../composables/useAuth.js'
-import { usePageTransition } from '../composables/usePageTransition.js'
+import { useAuth } from '@/composables/useAuth.js'
+import { usePageTransition } from '@/composables/usePageTransition.js'
 
+// ── 路由表 ──────────────────────────────────────────────
 const routes = [
+  // 首页
   {
     path: '/',
-    component: () => import('../views/LandingView.vue'),
-    meta: { layout: 'landing' },
+    component: () => import('@/views/HomeView.vue'),
+    meta: { layout: 'home' },
   },
+  // 认证页（登录 / 注册）
   {
     path: '/auth',
-    component: () => import('../components/AuthLayout.vue'),
+    component: () => import('@/views/auth/AuthLayout.vue'),
     redirect: '/auth/login',
     meta: { layout: 'auth' },
     children: [
-      { path: 'login',    component: () => import('../components/LoginView.vue'),    meta: { order: 0, layout: 'auth' } },
-      { path: 'register', component: () => import('../components/RegisterView.vue'), meta: { order: 1, layout: 'auth' } },
+      { path: 'login',    component: () => import('@/views/auth/LoginView.vue'), meta: { order: 0, layout: 'auth' } },
+      { path: 'register', component: () => import('@/views/auth/RegisterView.vue'), meta: { order: 1, layout: 'auth' } },
     ],
   },
+  // 工作台（需要登录，:view 参数决定子视图）
   {
     path: '/app/:view?/:subview?',
-    component: () => import('../views/WorkspaceView.vue'),
+    component: () => import('@/views/app/AppLayout.vue'),
     meta: { requiresAuth: true, layout: 'app' },
   },
+  // 兜底：未匹配路径重定向到工作台
   {
     path: '/:pathMatch(.*)*',
     redirect: '/app/workspace',
@@ -35,6 +47,7 @@ const router = createRouter({
   routes,
 })
 
+// ── 前置守卫：认证检查 + 过渡动画 ─────────────────────────
 router.beforeEach(async (to, from) => {
   const { currentUser, authChecked } = useAuth()
 
@@ -50,32 +63,33 @@ router.beforeEach(async (to, from) => {
     authChecked.value = true
   }
 
+  // 未登录 → 拦截到登录页
   if (to.meta.requiresAuth && !currentUser.value) {
     return '/auth/login'
   }
 
+  // 已登录 → 跳过认证页
   if (to.path.startsWith('/auth') && currentUser.value) {
     return '/app/workspace'
   }
 
-  // 跨布局跳转：先让 loading 完全进场，再放行路由
+  // 跨布局跳转：先让 loading 遮罩完全进场，再放行路由
   const fromLayout = from.meta.layout || ''
   const toLayout = to.meta.layout || ''
   if (fromLayout && toLayout && fromLayout !== toLayout) {
     const { show } = usePageTransition()
-    await show() // 等淡入动画结束
+    await show()
   }
 })
 
+// ── 后置守卫：跨布局过渡淡出 ──────────────────────────────
 router.afterEach(async (to, from) => {
   const fromLayout = from.meta.layout || ''
   const toLayout = to.meta.layout || ''
   if (fromLayout && toLayout && fromLayout !== toLayout) {
     const { hide } = usePageTransition()
-    // 等待 DOM 更新，确保新页面组件已挂载
-    await nextTick()
-    // 新页面渲染完成后再淡出，给一点缓冲时间
-    hide(500)
+    await nextTick()  // 等待 DOM 更新，确保新页面已挂载
+    hide(500)         // 延迟淡出，给渲染留缓冲时间
   }
 })
 
diff --git a/frontend/src/style.css b/frontend/src/style.css
index 5c0f7453..d8675371 100644
--- a/frontend/src/style.css
+++ b/frontend/src/style.css
@@ -11,10 +11,46 @@
     @apply inline-flex items-center justify-center gap-2 rounded-xl text-sm font-semibold shadow-md shadow-emerald-500/20 transition-all duration-300 active:scale-[0.97] motion-reduce:active:scale-100 focus:outline-none bg-gradient-to-r from-emerald-500 to-teal-600 text-white hover:from-emerald-400 hover:to-teal-500 hover:shadow-lg hover:shadow-emerald-500/30 focus:ring-4 focus:ring-emerald-500/30 disabled:cursor-not-allowed disabled:opacity-60 disabled:hover:shadow-md dark:shadow-none dark:hover:shadow-emerald-900/40;
   }
   .btn-secondary {
-    @apply inline-flex items-center justify-center gap-2 rounded-xl text-sm font-semibold shadow-md transition-all duration-300 active:scale-[0.97] motion-reduce:active:scale-100 focus:outline-none border border-slate-200 bg-white/80 backdrop-blur-sm text-slate-900 hover:bg-slate-50 hover:border-slate-300 focus:ring-4 focus:ring-slate-200/50 dark:border-slate-700/80 dark:bg-slate-900/80 dark:text-slate-200 dark:hover:bg-slate-800 dark:hover:border-slate-600 dark:focus:ring-slate-800/50;
+    @apply inline-flex items-center justify-center gap-2 rounded-xl text-sm font-semibold shadow-md transition-all duration-300 active:scale-[0.97] motion-reduce:active:scale-100 focus:outline-none border border-slate-200 bg-white/80 text-slate-900 hover:bg-slate-50 hover:border-slate-300 focus:ring-4 focus:ring-slate-200/50 dark:border-slate-700/80 dark:bg-slate-900/80 dark:text-slate-200 dark:hover:bg-slate-800 dark:hover:border-slate-600 dark:focus:ring-slate-800/50;
   }
 }
 
+/* ── 筛选 pill 按钮（Linear 风格） ── */
+.filter-pill {
+  display: inline-flex;
+  align-items: center;
+  padding: 4px 10px;
+  font-size: 13px;
+  font-weight: 500;
+  color: #8a8f98;
+  border-radius: 6px;
+  border: 1px solid rgba(255, 255, 255, 0.06);
+  background: rgba(255, 255, 255, 0.02);
+  transition: all 0.15s;
+  cursor: pointer;
+}
+.filter-pill:hover {
+  background: rgba(255, 255, 255, 0.04);
+  border-color: rgba(255, 255, 255, 0.1);
+  color: #d0d6e0;
+}
+.filter-pill--active {
+  background: rgba(129, 115, 223, 0.12);
+  border-color: rgba(129, 115, 223, 0.3);
+  color: rgb(145, 132, 235);
+}
+.filter-pill--active:hover {
+  background: rgba(129, 115, 223, 0.18);
+}
+
+/* ── 毛玻璃 active 按钮（Linear 风格，全局共用） ── */
+.brand-btn {
+  background: rgba(255, 255, 255, 0.05);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-top-color: rgba(255, 255, 255, 0.15);
+  border-bottom-color: rgba(255, 255, 255, 0.03);
+}
+
 html,
 body,
 #app {
diff --git a/frontend/src/utils.js b/frontend/src/utils.js
index f1c808d5..c3d278d0 100644
--- a/frontend/src/utils.js
+++ b/frontend/src/utils.js
@@ -23,11 +23,14 @@ export const isHtml = (s) => /<\/?(?:table|tr|td|th|thead|tbody)\b/i.test(s || '
 export const ALLOWED_HTML_TAGS = [
   'table', 'tr', 'td', 'th', 'thead', 'tbody',
   'p', 'br', 'span', 'b', 'i', 'em', 'strong', 'sub', 'sup',
+  'img',
 ]
 
 /** 使用 DOMPurify 过滤 HTML，仅保留白名单标签 */
-export const sanitizeHtml = (html) =>
-  DOMPurify.sanitize(html, { ALLOWED_TAGS: ALLOWED_HTML_TAGS })
+export const sanitizeHtml = (html) => {
+  const fixed = html.replace(/src="imgs\//g, 'src="/images/')
+  return DOMPurify.sanitize(fixed, { ALLOWED_TAGS: ALLOWED_HTML_TAGS, ALLOWED_ATTR: ['src', 'alt'] })
+}
 
 /** 从题目的 content_json 中提取纯文本摘要 */
 export const getQuestionSnippet = (q, maxLen = 0, fallback = '') => {
@@ -50,9 +53,9 @@ export const renderMarkdown = (text) => {
       ...ALLOWED_HTML_TAGS,
       'h1', 'h2', 'h3', 'h4', 'h5', 'h6',
       'ul', 'ol', 'li', 'code', 'pre', 'blockquote', 'hr',
-      'a',
+      'a', 'img',
     ],
-    ALLOWED_ATTR: ['href', 'target', 'rel'],
+    ALLOWED_ATTR: ['href', 'target', 'rel', 'src', 'alt'],
   })
 }
 
diff --git a/frontend/src/views/LandingView.vue b/frontend/src/views/HomeView.vue
similarity index 56%
rename from frontend/src/views/LandingView.vue
rename to frontend/src/views/HomeView.vue
index 12aca268..9099822a 100644
--- a/frontend/src/views/LandingView.vue
+++ b/frontend/src/views/HomeView.vue
@@ -1,13 +1,19 @@
 <script setup>
 import { ref, onMounted, onUnmounted, nextTick } from 'vue'
+import { useTheme } from '@/composables/useTheme.js'
 
-import LandingNav from '../components/landing/LandingNav.vue'
-import LandingHero from '../components/landing/LandingHero.vue'
-import LandingFeatures from '../components/landing/LandingFeatures.vue'
-import LandingWorkflow from '../components/landing/LandingWorkflow.vue'
-import LandingDemo from '../components/landing/LandingDemo.vue'
-import LandingCta from '../components/landing/LandingCta.vue'
-import LandingBackToTop from '../components/landing/LandingBackToTop.vue'
+const { initTheme, setTheme, isDark } = useTheme()
+
+// 首页固定深色：记住用户原始主题，离开时恢复
+let savedTheme = null
+
+import HomeHeader from '@/components/home/HomeHeader.vue'
+import HomeSideNav from '@/components/home/HomeSideNav.vue'
+import HomeHero from '@/views/home/HomeHero.vue'
+import HomeFeatures from '@/views/home/HomeFeatures.vue'
+import HomeWorkflow from '@/views/home/HomeWorkflow.vue'
+import HomeDemo from '@/views/home/HomeDemo.vue'
+import HomeFooter from '@/components/home/HomeFooter.vue'
 
 const SECTIONS = [
   { id: 'hero',     label: '首页' },
@@ -22,7 +28,6 @@ const activeSection = ref('hero')
 const backToTopVisible = ref(false)
 const navScrolled = ref(false)
 const indicatorTop = ref(44)
-
 // Cleanup holders
 let wheelUnlisten = null
 let scrollUnlisten = null
@@ -59,19 +64,31 @@ function getSectionTops() {
 }
 
 function onWheel(e) {
-  e.preventDefault()
   if (wheelLock) return
   const y = window.scrollY
   const tops = getSectionTops()
+  if (tops.length < 2) return
   const dir = e.deltaY > 0 ? 1 : -1
+
+  // 找到当前所在 section
   let cur = 0
   for (let i = tops.length - 1; i >= 0; i--) {
     if (y >= tops[i] - 10) { cur = i; break }
   }
-  const next = Math.max(0, Math.min(tops.length - 1, cur + dir))
-  if (next === cur) return
-  wheelLock = true
-  scrollToY(tops[next], 1000)
+
+  const lastIdx = tops.length - 1
+
+  // 只在 Hero(0) ↔ Features(1) 边界做 snap，其他正常滚动
+  const isHeroBoundary = (cur === 0 && dir === 1) || (cur === 1 && dir === -1 && y <= tops[1] + 50)
+
+  if (isHeroBoundary) {
+    e.preventDefault()
+    const next = Math.max(0, Math.min(lastIdx, cur + dir))
+    if (next === cur) return
+    wheelLock = true
+    scrollToY(tops[next], 800)
+  }
+  // 其他位置：不 preventDefault，浏览器正常滚动
 }
 
 // ── Scroll handler ──
@@ -182,11 +199,13 @@ function setupRevealObserver() {
 }
 
 onMounted(async () => {
-  // Landing page is always dark
-  document.documentElement.classList.add('dark')
+  initTheme()
+
+  // 首页强制深色模式
+  savedTheme = isDark.value ? 'dark' : 'light'
+  if (!isDark.value) setTheme(true)
 
-  // Set overflow-hidden on body for wheel snap
-  document.body.style.overflow = 'hidden'
+  // 不再锁定 body overflow，中间页面允许正常滚动
 
   await nextTick()
 
@@ -199,6 +218,7 @@ onMounted(async () => {
   window.addEventListener('scroll', onScrollThrottled, { passive: true })
   scrollUnlisten = () => window.removeEventListener('scroll', onScrollThrottled)
 
+  // Mousemove listener
   // Hero anim
   setTimeout(() => {
     document.querySelectorAll('.hero-anim').forEach(el => el.classList.add('is-visible'))
@@ -216,8 +236,8 @@ onMounted(async () => {
 })
 
 onUnmounted(() => {
-  // Restore body overflow
-  document.body.style.overflow = ''
+  // 离开首页时恢复用户原有主题
+  if (savedTheme === 'light') setTheme(false)
 
   if (wheelUnlisten) wheelUnlisten()
   if (scrollUnlisten) scrollUnlisten()
@@ -228,44 +248,71 @@ onUnmounted(() => {
 <template>
   <div class="min-h-screen bg-slate-50 text-slate-900 dark:bg-[#0A0A0F] dark:text-slate-300 selection:bg-blue-200 dark:selection:bg-indigo-500/30 page-enter">
 
-    <LandingNav
+    <!-- 🌟 真正的全局绝对底层固定背景 🌟 -->
+    <!-- 将背景移出任何可能带有 transform / overflow-hidden 的包裹层，确保 fixed 完美生效 -->
+    <HomeHeader
       :navScrolled="navScrolled"
       :activeSection="activeSection"
-      :indicatorTop="indicatorTop"
       :sections="SECTIONS"
       @scrollToSection="scrollToSectionSnap"
     />
+    <HomeSideNav
+      :activeSection="activeSection"
+      :sections="SECTIONS"
+      :showBackToTop="backToTopVisible"
+      @scrollToSection="scrollToSectionSnap"
+      @back-to-top="scrollToY(0, 1000)"
+    />
 
     <!-- 主内容包裹层 -->
     <div class="relative">
 
-      <LandingHero />
+      <HomeHero @scrollToSection="scrollToSectionSnap" />
 
-      <!-- ② 滚动叠盖层 -->
-      <div class="relative z-10 overflow-x-hidden">
+      <!-- ② 滚动叠盖层（实色背景完全遮住 Hero sticky） -->
+      <div class="relative z-10 overflow-hidden" style="background: #0A0A0F;">
 
-        <LandingFeatures />
+        <!-- 背景装饰：波纹纹理 -->
+        <div class="absolute inset-0 pointer-events-none z-0">
+          <div class="home-bg-topo"></div>
+        </div>
 
-        <LandingWorkflow />
+        <HomeFeatures />
 
-        <LandingDemo />
+        <HomeWorkflow />
 
-        <LandingCta />
+        <HomeDemo />
+
+        <HomeFooter />
+
+        <!-- 全局 Footer -->
+        <div class="border-t border-white/[0.06] py-4 relative z-10 bg-[#0A0A0F]/50 w-full mt-auto">
+          <div class="max-w-7xl mx-auto flex flex-col md:flex-row justify-between items-center gap-6 text-xs text-white/30 px-4">
+            <div class="flex items-center gap-2">
+              <div class="footer-logo relative overflow-hidden p-1.5 rounded-lg flex items-center justify-center">
+                <span class="footer-logo__grid absolute inset-0 pointer-events-none"></span>
+                <img src="/logo.svg" class="relative w-4 h-4 brightness-0 invert" alt="logo" />
+              </div>
+              <span class="font-semibold text-white/50 tracking-wide">智卷错题本</span>
+            </div>
+            <p class="text-center">© 2026 Intelligent Error Book Generation System. All rights reserved.</p>
+            <div class="flex gap-6 pr-12 md:pr-0">
+              <a href="#" class="hover:text-white/70 transition-colors">架构文档</a>
+              <a href="#" class="hover:text-white/70 transition-colors">隐私政策</a>
+              <a href="#" class="hover:text-white/70 transition-colors">联系我们</a>
+            </div>
+          </div>
+        </div>
 
       </div><!-- /滚动叠盖层 -->
 
     </div><!-- /主内容包裹层 -->
 
-    <LandingBackToTop
-      :visible="backToTopVisible"
-      @click="scrollToY(0, 1000)"
-    />
-
   </div>
 </template>
 
 <style>
-/* Landing page always uses dark mode class - set in onMounted */
+/* Home page always uses dark mode class - set in onMounted */
 
 @keyframes pageEnter {
   from { opacity: 0; }
@@ -298,18 +345,12 @@ onUnmounted(() => {
 }
 .animate-scan { animation: scan 3s ease-in-out infinite; }
 
-/* 动态玻璃态面板 */
+/* 动态玻璃态面板 — 统一为 brand-btn 风格 */
 .glass-panel {
-  background: rgba(255, 255, 255, 0.8);
-  backdrop-filter: blur(16px);
-  -webkit-backdrop-filter: blur(16px);
-  border: 1px solid rgba(0, 0, 0, 0.05);
-  box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05);
-}
-html.dark .glass-panel {
-  background: rgba(255, 255, 255, 0.03);
-  border: 1px solid rgba(255, 255, 255, 0.05);
-  box-shadow: none;
+  background: rgba(255, 255, 255, 0.05);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-top-color: rgba(255, 255, 255, 0.15);
+  border-bottom-color: rgba(255, 255, 255, 0.03);
 }
 
 /* 平滑滚动时补偿固定导航栏高度 */
@@ -343,4 +384,31 @@ html { scroll-padding-top: 80px; }
   opacity: 1;
   transform: translateY(0);
 }
+
+/* 滚动区背景装饰（与 Hero 一致） */
+.home-bg-topo {
+  position: absolute;
+  inset: 0;
+  opacity: 0.2;
+  background-image: url("data:image/svg+xml,%3Csvg viewBox='0 0 1000 1000' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='noiseFilter'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.005' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='matrix' values='1 0 0 0 0, 0 1 0 0 0, 0 0 1 0 0, 0 0 0 10 -4' /%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23noiseFilter)' fill='none' stroke='%23ffffff' stroke-width='1' opacity='0.3'/%3E%3Cpath d='M0,100 C200,300 300,0 500,100 C700,200 800,-100 1000,100 M0,200 C250,400 350,100 550,200 C750,300 850,0 1000,200 M0,300 C300,500 400,200 600,300 C800,400 900,100 1000,300 M0,400 C350,600 450,300 650,400 C850,500 950,200 1000,400 M0,500 C400,700 500,400 700,500 C900,600 1000,300 1000,500 M0,600 C450,800 550,500 750,600 C950,700 1000,400 1000,600 M0,700 C500,900 600,600 800,700 C1000,800 1000,500 1000,700 M0,800 C550,1000 650,700 850,800 C1000,900 1000,600 1000,800 M0,900 C600,1100 700,800 900,900 C1000,1000 1000,700 1000,900' stroke='%23ffffff' stroke-width='1' fill='none' opacity='0.15' /%3E%3C/svg%3E");
+  background-size: cover;
+  background-position: center;
+  mask-image: radial-gradient(ellipse 100% 100% at 50% 50%, black, transparent);
+  -webkit-mask-image: radial-gradient(ellipse 100% 100% at 50% 50%, black, transparent);
+}
+
+.footer-logo {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+
+.footer-logo__grid {
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+
 </style>
diff --git a/frontend/src/views/WorkspaceView.vue b/frontend/src/views/WorkspaceView.vue
deleted file mode 100644
index d3276b0d..00000000
--- a/frontend/src/views/WorkspaceView.vue
+++ /dev/null
@@ -1,1119 +0,0 @@
-<script setup>
-import { computed, nextTick, onBeforeUnmount, onMounted, provide, reactive, ref, watch } from 'vue'
-import { useRouter, useRoute } from 'vue-router'
-import { fileKey, typesetMath as _typesetMathEl } from '../utils.js'
-import * as api from '../api.js'
-import { useAuth } from '../composables/useAuth.js'
-import { usePageTransition } from '../composables/usePageTransition.js'
-// 注意：移除了 AppHeader，因为现在由左侧边栏接管了全局导航功能
-import StatusBar from '../components/StatusBar.vue'
-import StepIndicator from '../components/StepIndicator.vue'
-import FileList from '../components/FileList.vue'
-import FileUploader from '../components/FileUploader.vue'
-import QuestionList from '../components/QuestionList.vue'
-import ActionBar from '../components/ActionBar.vue'
-import SelectionPanel from '../components/SelectionPanel.vue'
-import SplitLoading from '../components/SplitLoading.vue'
-import ImageModal from '../components/ImageModal.vue'
-import ToastContainer from '../components/ToastContainer.vue'
-import Dashboard from '../components/Dashboard.vue'
-import ReviewView from '../components/ReviewView.vue'
-import ErrorBank from '../components/ErrorBank.vue'
-import ChatView from '../components/ChatView.vue'
-import SettingsView from '../components/SettingsView.vue'
-import SplitHistory from '../components/SplitHistory.vue'
-import GlassButton from '../components/GlassButton.vue'
-
-// ---- 认证状态 ----
-const { currentUser } = useAuth()
-const router = useRouter()
-const route = useRoute()
-
-const handleLogout = async () => {
-  try {
-    await fetch('/api/auth/logout', { method: 'POST' })
-  } catch (_) {}
-  currentUser.value = null
-  router.push('/auth')
-}
-
-const navigateToHome = () => {
-  document.body.style.transition = 'opacity 0.25s ease, transform 0.25s ease'
-  document.body.style.opacity = '0'
-  document.body.style.transform = 'translateY(-6px)'
-  setTimeout(() => { window.location.href = '/' }, 260)
-}
-
-// ---- 视图路由控制 ----
-const VIEW_TO_PATH = {
-  workspace: '/app/workspace',
-  workspace_review: '/app/workspace/review',
-  dashboard: '/app/dashboard',
-  review: '/app/review',
-  'error-bank': '/app/error-bank',
-  settings: '/app/settings',
-  'split-history': '/app/split-history',
-  chat: '/app/chat',
-}
-
-const WORKSPACE_VIEWS = new Set(['workspace', 'workspace_review', 'split-history'])
-
-const lastWorkspaceView = ref('workspace')
-
-const NAV_ITEMS = [
-  { id: 'workspace', label: '录题工作台', icon: 'fa-wand-magic-sparkles', match: (v) => WORKSPACE_VIEWS.has(v) },
-  { id: 'dashboard', label: '数据面板', icon: 'fa-chart-pie', match: (v) => v === 'dashboard' },
-  { id: 'error-bank', label: '错题库', icon: 'fa-database', match: (v) => v === 'error-bank' },
-]
-
-const currentView = computed({
-  get() {
-    const view = route.params.view || 'workspace'
-    const subview = route.params.subview
-    if (view === 'workspace' && subview === 'review') return 'workspace_review'
-    return view
-  },
-  set(view) {
-    const path = VIEW_TO_PATH[view]
-    if (path && route.path !== path) router.push(path)
-  },
-})
-
-watch(currentView, (v) => {
-  if (WORKSPACE_VIEWS.has(v)) lastWorkspaceView.value = v
-})
-
-const navIndicatorStyle = computed(() => {
-  const idx = NAV_ITEMS.findIndex(item => item.match(currentView.value))
-  if (idx === -1) return { opacity: 0, transform: 'translateY(0)' }
-  
-  // 精确计算：
-  // 1. 核心功能标题：text-xs (16px) + mb-2 (8px) = 24px
-  // 2. Flex Gap: gap-1.5 (6px)
-  // 3. 按钮高度：py-3 (12px * 2) + text-lg 图标高度 (28px) = 52px
-  const headerHeight = 24
-  const gap = 6
-  const itemHeight = 52
-  
-  const offset = headerHeight + gap
-  return {
-    opacity: 1,
-    transform: `translateY(${idx * (itemHeight + gap) + offset}px)`,
-    height: `${itemHeight}px`
-  }
-})
-
-// ---- 状态定义 ----
-const { loading: globalLoading } = usePageTransition()
-const theme = ref('dark')
-const applyTheme = (nextTheme) => {
-  theme.value = nextTheme === 'dark' ? 'dark' : 'light'
-  const root = document.documentElement
-  if (theme.value === 'dark') root.classList.add('dark')
-  else root.classList.remove('dark')
-}
-
-const toggleTheme = async (btnEl) => {
-  const nextTheme = theme.value === 'dark' ? 'light' : 'dark'
-  localStorage.setItem('theme', nextTheme)
-
-  const prefersReduce = window.matchMedia && window.matchMedia('(prefers-reduced-motion: reduce)').matches
-  const canTransition = !prefersReduce && typeof document.startViewTransition === 'function'
-  if (!canTransition || !btnEl) {
-    applyTheme(nextTheme)
-    return
-  }
-
-  const rect = btnEl.getBoundingClientRect()
-  const x = rect.left + rect.width / 2
-  const y = rect.top + rect.height / 2
-  const endRadius = Math.hypot(Math.max(x, window.innerWidth - x), Math.max(y, window.innerHeight - y))
-
-  const transition = document.startViewTransition(() => applyTheme(nextTheme))
-  try {
-    await transition.ready
-    const duration = 1000
-    const grow = [`circle(0px at ${x}px ${y}px)`, `circle(${endRadius}px at ${x}px ${y}px)`]
-    document.documentElement.animate(
-      { clipPath: grow },
-      { duration, easing: 'cubic-bezier(0.2, 0, 0, 1)', pseudoElement: '::view-transition-new(root)' },
-    )
-    document.documentElement.animate(
-      { opacity: [1, 0.98] },
-      { duration, easing: 'linear', pseudoElement: '::view-transition-old(root)' },
-    )
-    await transition.finished
-  } catch (_) {
-    applyTheme(nextTheme)
-  }
-}
-// ---- 系统状态 ----
-const statusLoading = ref(true)
-const systemStatus = ref(null)
-const statusError = ref('')
-const selectedModel = ref('')  // 选中的具体模型名称（如 "gpt-4o-mini"）
-
-const providerOptions = computed(() => {
-  const s = systemStatus.value
-  return s && s.available_models ? s.available_models : []
-})
-
-// 是否有可用模型
-const hasConfiguredModel = computed(() => providerOptions.value.some(p => p.configured))
-
-// 从 selectedModel 反查 provider
-const selectedProvider = computed(() => {
-  for (const p of providerOptions.value) {
-    if (p.models && p.models.includes(selectedModel.value)) return p.value
-  }
-  // 找不到时回退到首个已配置的 provider，而非硬编码 'openai'
-  const configured = providerOptions.value.find(p => p.configured)
-  return configured ? configured.value : (providerOptions.value[0]?.value ?? 'openai')
-})
-
-watch(systemStatus, (newVal) => {
-  if (newVal && newVal.available_models) {
-    const configured = newVal.available_models.find(m => m.configured)
-    if (configured) selectedModel.value = configured.default_model
-  }
-})
-
-const statusPills = computed(() => {
-  if (statusLoading.value) return [
-    { key: 'paddle', loading: true, label: 'PaddleOCR' },
-    { key: 'ensexam', loading: true, label: 'EnsExam' },
-    { key: 'langsmith', loading: true, label: 'LangSmith (未接入)' },
-  ]
-  const s = systemStatus.value
-  if (!s) return []
-  const pills = []
-  pills.push({ key: 'paddle', ok: !!s.paddleocr_configured, label: s.paddleocr_configured ? 'PaddleOCR' : 'PaddleOCR未配置' })
-  if (s.ensexam_configured) {
-    pills.push({ key: 'ensexam', ok: true, label: 'EnsExam已接入' })
-  }
-  pills.push(s.langsmith_enabled
-    ? { key: 'langsmith', ok: true, label: 'LangSmith追踪' }
-    : { key: 'langsmith', ok: false, label: 'LangSmith (未接入)', isPlaceholder: true }
-  )
-  return pills
-})
-
-const doFetchStatus = async () => {
-  statusLoading.value = true
-  statusError.value = ''
-  try {
-    systemStatus.value = await api.fetchStatus()
-  } catch (e) {
-    statusError.value = e instanceof Error ? e.message : String(e)
-  } finally {
-    statusLoading.value = false
-  }
-}
-
-// ---- 步骤 & Toast ----
-const step = ref(1)
-const toasts = ref([])
-let toastId = 0
-const pushToast = (type, message, timeout = 2600) => {
-  const id = ++toastId
-  toasts.value = [{ id, type, message }, ...toasts.value].slice(0, 5)
-  if (timeout > 0) window.setTimeout(() => { toasts.value = toasts.value.filter(t => t.id !== id) }, timeout)
-}
-provide('pushToast', pushToast)
-
-// ---- AI 辅导对话 ----
-const chatSessionId = ref(null)
-const chatQuestion = ref(null)
-const chatActive = ref(false)
-
-// 答案录入弹窗（AI 辅导前置）
-const answerModalOpen = ref(false)
-const answerModalTarget = ref(null)
-const answerModalText = ref('')
-const answerModalSaving = ref(false)
-
-const openChat = async (question) => {
-  chatQuestion.value = question
-  // 如果没有答案，先弹出答案录入弹窗
-  if (!question.answer) {
-    answerModalTarget.value = question
-    answerModalText.value = ''
-    answerModalOpen.value = true
-    return
-  }
-  await doOpenChatSession(question)
-}
-
-const doOpenChatSession = async (question) => {
-  try {
-    const sessions = await api.fetchChatSessions(question.id)
-    if (sessions.length) {
-      chatSessionId.value = sessions[0].id
-    } else {
-      const session = await api.createChat(question.id)
-      chatSessionId.value = session.id
-    }
-    chatActive.value = true
-    currentView.value = 'chat'
-  } catch (e) {
-    pushToast('error', '打开对话失败: ' + (e instanceof Error ? e.message : String(e)))
-  }
-}
-
-const saveAnswerAndChat = async () => {
-  if (!answerModalTarget.value || answerModalSaving.value) return
-  const text = answerModalText.value.trim()
-  if (!text) { pushToast('error', '请输入答案/解析内容'); return }
-  answerModalSaving.value = true
-  try {
-    await api.saveQuestionAnswer(answerModalTarget.value.id, text)
-    answerModalTarget.value.answer = text
-    answerModalOpen.value = false
-    pushToast('success', '答案已保存')
-    // 继续打开对话
-    await doOpenChatSession(answerModalTarget.value)
-  } catch (e) {
-    pushToast('error', '保存答案失败: ' + (e instanceof Error ? e.message : String(e)))
-  } finally {
-    answerModalSaving.value = false
-  }
-}
-
-const backToErrorBank = () => {
-  chatActive.value = false
-  chatSessionId.value = null
-  chatQuestion.value = null
-  currentView.value = 'error-bank'
-}
-
-// ---- 图片弹窗 ----
-const modalOpen = ref(false)
-const modalSrc = ref('')
-const modalScale = ref(1)
-const openModal = (src) => {
-  modalSrc.value = src || ''
-  modalScale.value = 1
-  modalOpen.value = !!src
-  if (src) document.body.style.overflow = 'hidden'
-}
-const closeModal = () => {
-  modalOpen.value = false
-  modalSrc.value = ''
-  modalScale.value = 1
-  document.body.style.overflow = ''
-}
-const onKeydown = (e) => { if (e.key === 'Escape' && modalOpen.value) closeModal() }
-
-// ---- 上传状态 ----
-const uploadBusy = ref(false)
-const uploadReady = ref(false)
-const splitting = ref(false)
-const splitCompleted = ref(false)
-const eraseEnabled = ref(true)
-
-const pendingFiles = reactive([])
-const fileProgress = reactive({})
-const waitingKeys = reactive(new Set())
-const uploadQueue = reactive([])
-let activeXhr = null
-let fakeProgressTimer = null
-let fakeProgressKeys = []
-
-const splitEnabled = computed(() => !splitting.value && !splitCompleted.value && uploadReady.value && !uploadBusy.value && hasConfiguredModel.value)
-const exportEnabled = computed(() => splitCompleted.value && selectedIds.size > 0)
-
-const stopFakeProgress = () => {
-  if (fakeProgressTimer) {
-    window.clearInterval(fakeProgressTimer)
-    fakeProgressTimer = null
-  }
-  fakeProgressKeys = []
-}
-const setProgress = (key, p) => { fileProgress[key] = Math.max(0, Math.min(100, Number(p) || 0)) }
-
-const startFakeProgress = (keys) => {
-  stopFakeProgress()
-  fakeProgressKeys = Array.from(keys || [])
-  const tick = () => {
-    if (!uploadBusy.value) { stopFakeProgress(); return }
-    for (const key of fakeProgressKeys) {
-      const current = Number(fileProgress[key] || 0)
-      const cap = 82
-      if (current >= cap) continue
-      let inc = current < 55 ? 1 + Math.random() * 3 : current < 75 ? 0.4 + Math.random() * 1.1 : 0.08 + Math.random() * 0.25
-      setProgress(key, Math.min(cap, current + inc))
-    }
-  }
-  tick()
-  fakeProgressTimer = window.setInterval(tick, 360)
-}
-
-const enqueueUpload = (files) => {
-  const list = Array.from(files || [])
-  if (!list.length) return
-  if (splitCompleted.value || splitting.value) { pushToast('error', '已分割完成，请先重新开始'); return }
-  for (const f of list) {
-    const k = fileKey(f)
-    if (pendingFiles.some(x => x.key === k)) continue
-    pendingFiles.push({ key: k, file: f })
-    setProgress(k, 0)
-  }
-  if (uploadBusy.value) {
-    for (const f of list) waitingKeys.add(fileKey(f))
-    uploadQueue.push(list)
-    return
-  }
-  handleUpload(list)
-}
-
-const pumpUploadQueue = () => {
-  if (uploadBusy.value || !uploadQueue.length) return
-  const next = uploadQueue.shift()
-  if (next && next.length) handleUpload(next)
-}
-const removePendingFile = async (key) => {
-  if (!key || splitting.value || splitCompleted.value) return
-  if (uploadBusy.value || uploadReady.value) { await doCancelFile(key); return }
-  const idx = pendingFiles.findIndex(x => x.key === key)
-  if (idx >= 0) pendingFiles.splice(idx, 1)
-  delete fileProgress[key]
-  waitingKeys.delete(key)
-  for (let i = uploadQueue.length - 1; i >= 0; i--) {
-    uploadQueue[i] = (uploadQueue[i] || []).filter(f => fileKey(f) !== key)
-    if (!uploadQueue[i].length) uploadQueue.splice(i, 1)
-  }
-}
-
-const doCancelFile = async (key) => {
-  try {
-    if (fakeProgressKeys.length) fakeProgressKeys = fakeProgressKeys.filter(k => k !== key)
-    if (!fakeProgressKeys.length) stopFakeProgress()
-    const data = await api.cancelFile(key)
-    const idx = pendingFiles.findIndex(x => x.key === key)
-    if (idx >= 0) pendingFiles.splice(idx, 1)
-    delete fileProgress[key]
-    waitingKeys.delete(key)
-    questions.value = []
-    selectedIds.clear()
-    splitCompleted.value = false
-    if (!pendingFiles.length) {
-      uploadReady.value = false
-      step.value = 1
-    } else {
-      step.value = 3
-    }
-    pushToast('success', data.message || '已撤销')
-    if (!pendingFiles.length && activeXhr) {
-      try { activeXhr.abort() } catch (_) {}
-    }
-  } catch (_) { pushToast('error', '撤销失败: 网络错误') }
-}
-
-const handleUpload = (files) => {
-  const uploadFiles = Array.from(files || []).filter(f => pendingFiles.some(x => x.key === fileKey(f)))
-  if (!uploadFiles.length) return
-  uploadBusy.value = true
-  step.value = 2
-  const keys = uploadFiles.map(f => fileKey(f))
-  for (const k of keys) waitingKeys.delete(k)
-  startFakeProgress(keys)
-
-  const formData = new FormData()
-  for (const f of uploadFiles) {
-    formData.append('files', f)
-    formData.append('file_key', fileKey(f))
-  }
-
-  activeXhr = api.uploadFiles(formData, {
-    onProgress: (ratio) => {
-      const pct = Math.max(0, Math.min(95, ratio * 95))
-      for (const k of keys) {
-        if (pendingFiles.some(x => x.key === k)) setProgress(k, Math.max(fileProgress[k] || 0, pct))
-      }
-    },
-    onSuccess: () => {
-      stopFakeProgress()
-      uploadBusy.value = false
-      activeXhr = null
-      for (const k of keys) {
-        if (pendingFiles.some(x => x.key === k)) setProgress(k, 100)
-      }
-      uploadReady.value = pendingFiles.length > 0
-      step.value = pendingFiles.length > 0 ? 3 : 1
-      pushToast('success', `上传成功！本次新增 ${keys.length} 个文件，点击"开始分割题目"开始处理`)
-      pumpUploadQueue()
-    },
-    onError: (msg) => {
-      stopFakeProgress()
-      uploadBusy.value = false
-      activeXhr = null
-      pushToast('error', msg)
-      pumpUploadQueue()
-    },
-    onAbort: () => {
-      stopFakeProgress()
-      uploadBusy.value = false
-      activeXhr = null
-      pumpUploadQueue()
-    },
-  })
-}
-
-// ---- 题目 ----
-const questions = ref([])
-const selectedIds = reactive(new Set())
-const questionListRef = ref(null)
-const errorBankRef = ref(null)
-
-const toggleQuestion = (id) => { selectedIds.has(id) ? selectedIds.delete(id) : selectedIds.add(id) }
-const selectAll = () => { for (const q of questions.value) selectedIds.add(q.question_id) }
-const deselectAll = () => { selectedIds.clear() }
-const reorderQuestions = (oldIndex, newIndex) => {
-  const arr = questions.value.slice()
-  const [moved] = arr.splice(oldIndex, 1)
-  arr.splice(newIndex, 0, moved)
-  questions.value = arr
-}
-
-const typesetMath = async () => {
-  await nextTick()
-  const el = questionListRef.value?.questionsBoxEl
-  await _typesetMathEl(el || undefined)
-}
-const doSplit = async () => {
-  if (!uploadReady.value || splitting.value || splitCompleted.value) return
-  splitting.value = true
-  step.value = 3
-  pushToast('info', '正在调用AI分割题目，请稍候...', 1800)
-  try {
-    const data = await api.splitQuestions(selectedProvider.value, selectedModel.value, { erase: eraseEnabled.value })
-    questions.value = data.questions || []
-    selectedIds.clear()
-    if (data.warnings && data.warnings.length) {
-      for (const w of data.warnings) pushToast('warning', w, 6000)
-    }
-    if (questions.value.length > 0) {
-      splitCompleted.value = true
-      step.value = 4
-      if (!data.warnings || !data.warnings.length) {
-        pushToast('success', `成功分割 ${questions.value.length} 道题目`)
-      }
-      await typesetMath()
-      // 分割完成后自动翻页到核对页面（仅在用户仍在工作台时跳转）
-      setTimeout(() => {
-        if (currentView.value === 'workspace') {
-          currentView.value = 'workspace_review'
-        }
-      }, 800)
-    }
-  } catch (e) {
-    pushToast('error', '分割失败: ' + (e instanceof Error ? e.message : String(e)))
-  } finally {
-    splitting.value = false
-  }
-}
-
-const doExport = async () => {
-  if (!selectedIds.size) { pushToast('error', '请至少选择一道题目！'); return }
-  try {
-    const data = await api.exportQuestions(Array.from(selectedIds))
-    step.value = 5
-    pushToast('success', `错题本导出成功！已保存到: ${data.output_path}`)
-    let filename = 'wrongbook.md'
-    if (data.output_path) {
-      const parts = String(data.output_path).split(/[/\\]/)
-      const last = parts[parts.length - 1]
-      if (last) filename = last
-    }
-    const a = document.createElement('a')
-    a.href = `/download/${encodeURIComponent(filename)}?t=${Date.now()}`
-    a.download = filename
-    a.style.display = 'none'
-    document.body.appendChild(a)
-    a.click()
-    a.remove()
-  } catch (e) {
-    pushToast('error', '导出失败: ' + (e instanceof Error ? e.message : String(e)))
-  }
-}
-
-const doSaveToDb = async () => {
-  if (!selectedIds.size) { pushToast('error', '请至少选择一道题目！'); return }
-  try {
-    // 收集已录入的答案数据一并传给后端
-    const answers = questions.value
-      .filter(q => selectedIds.has(q.question_id) && (q.answer || q.user_answer))
-      .map(q => ({ question_id: q.question_id, answer: q.answer || '', user_answer: q.user_answer || '' }))
-    const data = await api.saveToDb(Array.from(selectedIds), answers)
-    pushToast('success', data.message || '已导入错题库')
-    errorBankRef.value?.refresh()
-  } catch (e) {
-    pushToast('error', '导入失败: ' + (e instanceof Error ? e.message : String(e)))
-  }
-}
-
-const handleLoadRecord = (qs, record) => {
-  questions.value = qs || []
-  selectedIds.clear()
-  splitCompleted.value = true
-  step.value = 4
-  currentView.value = 'workspace_review'
-  pushToast('success', `已加载「${record?.subject || '历史记录'}」的 ${qs.length} 道题目`)
-  nextTick(() => typesetMath())
-}
-
-const doReset = () => {
-  uploadBusy.value = false
-  uploadReady.value = false
-  splitting.value = false
-  splitCompleted.value = false
-  pendingFiles.splice(0, pendingFiles.length)
-  for (const k of Object.keys(fileProgress)) delete fileProgress[k]
-  waitingKeys.clear()
-  uploadQueue.splice(0, uploadQueue.length)
-  questions.value = []
-  selectedIds.clear()
-  const configured = providerOptions.value.find(m => m.configured)
-  selectedModel.value = configured ? configured.default_model : ''
-  step.value = 1
-  pushToast('success', '已重置')
-}
-
-watch(currentView, async (newView) => {
-  if (newView === 'workspace_review') {
-    // 等待翻页动画结束
-    await nextTick()
-    setTimeout(() => {
-      questionListRef.value?.triggerTypeset?.()
-    }, 650)
-  }
-})
-
-// ---- 页面加载动画 ----
-const pageLoading = ref(true)
-
-// ---- 生命周期 ----
-onMounted(() => {
-  applyTheme('dark')
-  document.addEventListener('keydown', onKeydown)
-
-  // 刷新时如果落在 /workspace/review 但没有数据，重定向回上传页
-  if (currentView.value === 'workspace_review' && !splitCompleted.value) {
-    router.replace('/app/workspace')
-  }
-  
-  // 延迟系统状态检查，直到入场加载动画完全结束
-  // 这能确保在动画过程中不会因为网络请求导致掉帧，且视觉上更连贯
-  setTimeout(() => {
-    pageLoading.value = false
-    
-    // 如果全局 loading 已经结束（说明动画已经淡出或不需要淡出），开始检查
-    if (!globalLoading.value) {
-      doFetchStatus()
-    } else {
-      // 否则，监听全局 loading 状态，待其变为 false（即 AppLoading 彻底消失）后触发
-      const unwatch = watch(globalLoading, (val) => {
-        if (!val) {
-          doFetchStatus()
-          unwatch()
-        }
-      })
-    }
-  }, 2000)
-})
-
-onBeforeUnmount(() => {
-  stopFakeProgress()
-  document.removeEventListener('keydown', onKeydown)
-})
-</script>
-
-<template>
-  <div class="flex h-screen w-full overflow-hidden bg-slate-50 font-sans text-slate-900 dark:bg-[#05050A] dark:text-slate-300">
-
-    <Transition name="ws-loading-fade">
-      <div v-if="pageLoading" class="fixed inset-0 z-[200] flex flex-col items-center justify-center gap-8 bg-[#05050A]">
-        <div class="relative">
-          <div class="absolute inset-0 rounded-2xl bg-indigo-500/40 blur-xl"></div>
-          <div class="relative bg-gradient-to-br from-indigo-500 to-indigo-700 p-3.5 rounded-2xl shadow-lg border border-white/10">
-            <img src="/logo.svg" class="w-9 h-9 brightness-0 invert" alt="logo" />
-          </div>
-        </div>
-        <div class="w-48">
-          <div class="h-0.5 w-full rounded-full bg-white/10 overflow-hidden">
-            <div class="h-full rounded-full bg-gradient-to-r from-indigo-500 to-blue-400 ws-loading-bar"></div>
-          </div>
-        </div>
-      </div>
-    </Transition>
-    <!-- ================== 全局固定背景光晕 (支持长页面滚动) ================== -->
-    <div class="pointer-events-none fixed inset-0 z-0 overflow-hidden">
-      <div class="animate-blob absolute -top-[10%] left-[-10%] h-[50vw] w-[50vw] rounded-full bg-blue-400/[0.12] mix-blend-multiply blur-[120px] dark:bg-indigo-500/[0.15] dark:mix-blend-screen"></div>
-      <div class="animate-blob animation-delay-4000 absolute -bottom-[10%] right-[-10%] h-[45vw] w-[45vw] rounded-full bg-indigo-300/[0.15] mix-blend-multiply blur-[100px] dark:bg-fuchsia-500/[0.12] dark:mix-blend-screen"></div>
-      <div class="animate-blob animation-delay-2000 absolute left-[15%] top-[25%] h-[35vw] w-[35vw] rounded-full bg-cyan-200/[0.12] mix-blend-multiply blur-[110px] dark:bg-blue-500/[0.10] dark:mix-blend-screen"></div>
-    </div>
-
-    <!-- ================== PC端：左侧边栏导航 ================== -->
-    <aside class="hidden w-64 flex-col justify-between border-r border-slate-200 bg-white md:flex dark:border-white/5 dark:bg-[#0A0A0F]/80 z-20">
-      <div>
-        <!-- Logo 标题区 -->
-        <div class="flex h-20 items-center gap-2 border-b border-slate-100 px-4 dark:border-white/5">
-          <button @click="navigateToHome" class="flex flex-1 min-w-0 items-center gap-3 rounded-xl px-2 py-1.5 hover:bg-slate-100 dark:hover:bg-white/5 transition-colors" title="返回介绍页">
-            <div class="relative flex h-10 w-10 shrink-0 items-center justify-center rounded-xl bg-gradient-to-br from-blue-500 to-indigo-600 shadow-lg shadow-blue-500/30 dark:shadow-indigo-500/20">
-              <img src="/logo.svg" class="w-6 h-6 brightness-0 invert relative z-10" alt="logo" />
-              <div class="absolute inset-0 animate-pulse rounded-xl bg-blue-400/20 blur-md"></div>
-            </div>
-            <span class="text-xl font-extrabold tracking-tight text-transparent bg-clip-text bg-gradient-to-r from-blue-700 to-indigo-700 dark:from-white dark:to-indigo-200">
-              智卷系统
-            </span>
-          </button>
-          <button
-            @click="(e) => toggleTheme(e.currentTarget)"
-            class="hidden"
-            title="切换主题"
-          >
-            <i class="fa-solid fa-sun text-[18px] hidden dark:block"></i>
-            <i class="fa-solid fa-moon text-[18px] block dark:hidden"></i>
-          </button>
-        </div>
-
-        <!-- 视图切换菜单 -->
-        <nav class="mt-6 flex flex-col gap-1.5 px-4 relative">
-          <div class="mb-2 px-3 text-xs font-bold uppercase tracking-[0.2em] text-slate-400 dark:text-slate-500">核心功能</div>
-
-          <!-- 悬浮指示器背景 -->
-          <div
-            class="absolute left-4 right-4 z-0 rounded-xl bg-blue-600 shadow-lg shadow-blue-600/20 transition-all duration-300 ease-[cubic-bezier(0.34,1.56,0.64,1)] dark:bg-indigo-500 dark:shadow-indigo-500/20"
-            :style="navIndicatorStyle"
-          ></div>
-
-          <!-- 激活指示小圆点 -->
-          <div
-            class="absolute right-6 z-20 h-1.5 w-1.5 rounded-full bg-white/80 transition-all duration-300 ease-[cubic-bezier(0.34,1.56,0.64,1)] pointer-events-none"
-            :style="{
-              opacity: navIndicatorStyle.opacity,
-              transform: `translateY(calc(${navIndicatorStyle.transform.match(/translateY\(([^)]+)\)/)[1]} + 25.25px))`,
-            }"
-          ></div>
-
-          <button
-            v-for="item in NAV_ITEMS"
-            :key="item.id"
-            @click="currentView = (item.id === 'workspace' ? lastWorkspaceView : item.id)"
-            class="group relative z-10 flex items-center gap-3 rounded-xl px-4 py-3 text-sm font-bold transition-all duration-200"
-            :class="item.match(currentView)
-              ? 'text-white'
-              : 'text-slate-600 hover:bg-slate-100/50 hover:text-blue-600 dark:text-slate-400 dark:hover:bg-white/5 dark:hover:text-indigo-300'"
-          >
-            <i class="w-5 text-center text-lg transition-transform group-hover:scale-110" :class="[item.icon, 'fa-solid']"></i>
-            <span>{{ item.label }}</span>
-          </button>
-
-          <button
-            disabled
-            class="group relative flex items-center gap-3 rounded-xl px-4 py-3 text-sm font-bold cursor-not-allowed opacity-40 text-slate-400 dark:text-slate-600"
-          >
-            <i class="fa-solid fa-clock-rotate-left w-5 text-center text-lg"></i>
-            <span>刷题</span>
-          </button>
-        </nav>
-      </div>
-
-      <!-- 底部控制与返回栏 -->
-      <div class="space-y-1.5 border-t border-slate-100 p-4 dark:border-white/5">
-        <!-- 用户信息 -->
-        <div class="flex items-center gap-2 rounded-2xl px-3 py-2 mb-2 bg-white/70 backdrop-blur-xl border border-slate-200/60 dark:bg-white/[0.06] dark:backdrop-blur-xl dark:border-white/10">
-          <div class="h-8 w-8 shrink-0 rounded-full bg-gradient-to-br from-blue-500 to-indigo-600 dark:from-indigo-400 dark:to-indigo-600 flex items-center justify-center text-white text-sm font-extrabold shadow-sm">
-            {{ currentUser?.username?.[0]?.toUpperCase() ?? '?' }}
-          </div>
-          <div class="flex-1 min-w-0">
-            <p class="text-sm font-semibold text-slate-700 dark:text-slate-200 truncate leading-tight">{{ currentUser?.username }}</p>
-            <p class="text-xs text-slate-400 dark:text-slate-500 truncate leading-tight">{{ currentUser?.email }}</p>
-          </div>
-          <button
-            @click="handleLogout"
-            title="退出登录"
-            class="shrink-0 h-7 w-7 flex items-center justify-center rounded-lg text-slate-400 hover:text-rose-500 hover:bg-rose-50/80 dark:hover:bg-rose-500/10 transition-all"
-          >
-            <i class="fas fa-right-from-bracket text-sm"></i>
-          </button>
-        </div>
-
-        <button
-          @click="currentView = 'settings'"
-          class="flex w-full items-center gap-3 rounded-xl px-4 py-3 text-sm font-semibold transition-all duration-200"
-          :class="currentView === 'settings' ? 'bg-white/70 backdrop-blur-xl border border-blue-200/60 text-blue-600 shadow-sm dark:bg-white/10 dark:backdrop-blur-xl dark:border-indigo-500/30 dark:text-indigo-300 dark:shadow-[0_0_15px_rgba(99,102,241,0.15)]' : 'text-slate-600 hover:bg-white/60 hover:backdrop-blur-xl hover:text-blue-600 dark:text-slate-400 dark:hover:bg-white/5 dark:hover:text-indigo-300'"
-        >
-          <i class="fa-solid fa-gear w-5 text-center text-lg"></i>
-          系统设置
-        </button>
-
-      </div>
-    </aside>
-
-    <!-- ================== 移动端：底部 Tab 导航栏 ================== -->
-    <nav class="fixed bottom-0 left-0 right-0 z-50 border-t border-slate-200 bg-white/90 pb-2 pt-2 backdrop-blur-xl md:hidden dark:border-white/10 dark:bg-[#0A0A0F]/90">
-      <div class="flex justify-around">
-        <button @click="currentView = lastWorkspaceView" class="flex flex-col items-center p-2" :class="WORKSPACE_VIEWS.has(currentView) ? 'text-blue-600 dark:text-indigo-400' : 'text-slate-500 dark:text-slate-400'">
-          <i class="fa-solid fa-file-arrow-up text-lg"></i>
-          <span class="mt-1 text-xs font-bold">录题</span>
-        </button>
-        <button disabled class="flex flex-col items-center p-2 cursor-not-allowed opacity-40 text-slate-400 dark:text-slate-600">
-          <i class="fa-solid fa-clock-rotate-left text-lg"></i>
-          <span class="mt-1 text-xs font-bold">刷题</span>
-        </button>
-        <button @click="currentView = 'dashboard'" class="flex flex-col items-center p-2" :class="currentView === 'dashboard' ? 'text-blue-600 dark:text-indigo-400' : 'text-slate-500 dark:text-slate-400'">
-          <i class="fa-solid fa-chart-pie text-lg"></i>
-          <span class="mt-1 text-xs font-bold">数据面板</span>
-        </button>
-        <button @click="currentView = 'error-bank'" class="flex flex-col items-center p-2" :class="currentView === 'error-bank' ? 'text-blue-600 dark:text-indigo-400' : 'text-slate-500 dark:text-slate-400'">
-          <i class="fa-solid fa-layer-group text-lg"></i>
-          <span class="mt-1 text-xs font-bold">错题本</span>
-        </button>
-        <button @click="currentView = 'settings'" class="flex flex-col items-center p-2" :class="currentView === 'settings' ? 'text-blue-600 dark:text-indigo-400' : 'text-slate-500 dark:text-slate-400'">
-          <i class="fa-solid fa-sliders text-lg"></i>
-          <span class="mt-1 text-xs font-bold">设置</span>
-        </button>
-        <button @click="(e) => toggleTheme(e.currentTarget)" class="hidden">
-          <i class="fa-solid mb-1 text-xl" :class="theme === 'dark' ? 'fa-sun text-indigo-400' : 'fa-moon'"></i>
-          <span class="text-xs font-bold">主题</span>
-        </button>
-      </div>
-    </nav>
-
-    <!-- ================== 右侧主内容区 ================== -->
-    <main class="relative z-10 flex-1 overflow-hidden pb-20 md:pb-0">
-
-      <Transition name="view-fade" mode="out-in">
-        <!-- 视图 1：录题工作台（分两页：上传解析页 / 题目核对页） -->
-        <div v-if="currentView === 'workspace' || currentView === 'workspace_review'" key="workspace" class="relative h-full flex flex-col overflow-hidden">
-          <div class="container relative z-10 mx-auto flex h-full min-h-0 max-w-6xl flex-col px-4 py-4 sm:px-8 sm:py-6">
-            <Transition name="flip" mode="out-in">
-              <!-- 第一页：录题与分析 -->
-              <div v-if="currentView === 'workspace'" key="upload" class="flex flex-1 flex-col min-h-0">
-                <div class="mb-4 flex flex-col items-start gap-2 pl-2 sm:pl-0 md:flex-row md:items-center md:justify-between shrink-0">
-                  <div>
-                    <h2 class="text-2xl font-extrabold tracking-tight text-slate-900 sm:text-3xl dark:text-white">
-                      智能录入与分析
-                    </h2>
-                    <div class="mt-2 flex items-center">
-                      <div class="relative flex items-center gap-2 py-1 text-xs font-black tracking-widest text-blue-600/80 dark:text-indigo-300/80">
-                        <div class="relative flex h-4 w-4 items-center justify-center">
-                          <i class="fa-solid fa-bolt-lightning absolute animate-pulse text-xs text-indigo-500 dark:text-indigo-400"></i>
-                          <div class="absolute h-full w-full animate-ping rounded-full bg-indigo-400/10 dark:bg-indigo-400/10"></div>
-                        </div>
-
-                        <span class="relative z-10 uppercase pb-0.5">
-                          新一代 AI 错题处理架构 <span class="ml-1 font-extrabold text-blue-600 dark:text-indigo-300">V2.0</span>
-                        </span>
-                      </div>
-                    </div>
-                  </div>
-                  <GlassButton icon="fa-clock-rotate-left" @click="currentView = 'split-history'">
-                    分割历史
-                  </GlassButton>
-                </div>
-
-                <div class="main-content relative flex flex-1 flex-col min-h-0 bg-transparent">
-                  <StatusBar
-                    class="border-b border-slate-200/60 pb-6 dark:border-white/5"
-                    :status-loading="statusLoading"
-                    :status-error="statusError"
-                    :status-pills="statusPills"
-                    :provider-options="providerOptions"
-                    :selected-model="selectedModel"
-                    :disabled="splitting || splitCompleted"
-                    :no-models="!hasConfiguredModel"
-                    @update:selected-model="(v) => selectedModel = v"
-                  />
-
-                  <!-- 功能开关 -->
-                  <div class="flex items-center gap-4 border-b border-slate-200/60 py-4 dark:border-white/5">
-                    <label class="flex cursor-pointer items-center gap-2" @click="eraseEnabled = !eraseEnabled">
-                      <div class="relative h-6 w-10 rounded-full transition-colors" :class="eraseEnabled ? 'bg-indigo-500' : 'bg-slate-300 dark:bg-slate-600'">
-                        <div class="absolute top-1 h-4 w-4 rounded-full bg-white shadow-sm transition-transform" :class="eraseEnabled ? 'translate-x-5' : 'translate-x-1'"></div>
-                      </div>
-                      <span class="text-sm font-bold text-slate-700 dark:text-slate-300">擦除手写字迹</span>
-                      <span class="text-xs text-slate-400 dark:text-slate-500">上传后自动擦除图片中的手写笔迹</span>
-                    </label>
-                  </div>
-
-                  <div class="flex-1 overflow-y-auto pr-2 custom-scrollbar flex flex-col space-y-8 py-6">
-                    <StepIndicator :step="step" class="border-b border-slate-200/60 pb-8 dark:border-white/5" />
-
-                    <FileList
-                      :pending-files="pendingFiles"
-                      :file-progress="fileProgress"
-                      :waiting-keys="waitingKeys"
-                      :upload-busy="uploadBusy"
-                      :upload-ready="uploadReady"
-                      :splitting="splitting"
-                      :split-completed="splitCompleted"
-                      @remove-file="removePendingFile"
-                    />
-
-                    <FileUploader
-                      class="transition-all duration-500"
-                      :class="questions.length === 0 ? 'flex-1' : 'border-b border-slate-200/60 pb-8 dark:border-white/5'"
-                      :pending-files="pendingFiles"
-                      :file-progress="fileProgress"
-                      :waiting-keys="waitingKeys"
-                      :upload-busy="uploadBusy"
-                      :upload-ready="uploadReady"
-                      :splitting="splitting"
-                      :split-completed="splitCompleted"
-                      :expand="questions.length === 0"
-                      :disabled="!hasConfiguredModel"
-                      @upload="enqueueUpload"
-                      @remove-file="removePendingFile"
-                    />
-                  </div>
-
-                  <ActionBar
-                    class="shrink-0 border-t border-slate-200/60 pt-6 dark:border-white/5"
-                    :split-enabled="splitEnabled"
-                    :export-enabled="false"
-                    :splitting="splitting"
-                    :split-completed="splitCompleted"
-                    @split="doSplit"
-                  />
-
-                </div>
-              </div>
-
-              <!-- 第二页：解析结果核对 -->
-              <div v-else-if="currentView === 'workspace_review'" key="review" class="flex flex-1 flex-col overflow-hidden">
-                <div class="mb-4 flex items-center justify-between pl-2 sm:pl-0 shrink-0">
-                  <div class="flex items-center gap-4">
-                    <button
-                      @click="() => { doReset(); currentView = 'workspace' }"
-                      class="group flex h-10 w-10 items-center justify-center rounded-xl border border-slate-200/60 bg-white/60 text-slate-500 backdrop-blur-md transition-all hover:border-blue-500/50 hover:bg-white hover:text-blue-600 dark:border-white/10 dark:bg-white/5 dark:text-slate-400 dark:hover:border-indigo-500/50 dark:hover:text-indigo-300"
-                    >
-                      <i class="fa-solid fa-arrow-left-long transition-transform group-hover:-translate-x-1"></i>
-                    </button>
-                    <div>
-                      <h2 class="text-2xl font-extrabold tracking-tight text-slate-900 sm:text-3xl dark:text-white">
-                        题目数据核对
-                      </h2>
-                      <p class="text-xs font-bold text-slate-400 dark:text-slate-500 mt-1">请确认解析结果的准确性并进行导出或存档</p>
-                    </div>
-                  </div>
-                </div>
-
-                <div class="flex-1 overflow-y-auto pr-2 custom-scrollbar py-2 pb-24">
-                  <QuestionList
-                    ref="questionListRef"
-                    :questions="questions"
-                    :selected-ids="selectedIds"
-                    @toggle-select="toggleQuestion"
-                    @select-all="selectAll"
-                    @deselect-all="deselectAll"
-                    @open-image="openModal"
-                    @reorder="reorderQuestions"
-                  />
-                </div>
-
-              </div>
-            </Transition>
-
-          </div>
-        </div>
-
-        <!-- 视图 2：待复习 -->
-        <div v-else-if="currentView === 'review'" key="review_view" class="h-full">
-          <ReviewView
-            :theme="theme"
-            :visible="currentView === 'review'"
-            @go-workspace="currentView = 'workspace'"
-            @push-toast="pushToast"
-            @open-image="openModal"
-            @start-chat="openChat"
-          />
-        </div>
-
-        <!-- 视图 3：数据面板 -->
-        <div v-else-if="currentView === 'dashboard'" key="dashboard_view" class="h-full">
-          <Dashboard
-            :theme="theme"
-            :visible="currentView === 'dashboard'"
-            @go-workspace="currentView = 'workspace'"
-            @push-toast="pushToast"
-          />
-        </div>
-
-        <!-- 视图 4：错题库 -->
-        <div v-else-if="currentView === 'error-bank'" key="error_bank_view" class="h-full">
-          <ErrorBank
-            ref="errorBankRef"
-            :theme="theme"
-            :visible="currentView === 'error-bank'"
-            @go-workspace="currentView = 'workspace'"
-            @push-toast="pushToast"
-            @open-image="openModal"
-            @start-chat="openChat"
-          />
-        </div>
-
-        <!-- 视图 5：分割历史 -->
-        <div v-else-if="currentView === 'split-history'" key="split_history_view" class="h-full">
-          <SplitHistory
-            :theme="theme"
-            :visible="currentView === 'split-history'"
-            @push-toast="pushToast"
-            @open-image="openModal"
-            @load-record="handleLoadRecord"
-            @go-workspace="currentView = splitCompleted ? 'workspace_review' : 'workspace'"
-          />
-        </div>
-
-        <!-- 视图 6：系统设置 -->
-        <div v-else-if="currentView === 'settings'" key="settings_view" class="h-full">
-          <SettingsView
-            :visible="currentView === 'settings'"
-            @saved="doFetchStatus"
-          />
-        </div>
-
-        <!-- 视图 7：AI 辅导对话 -->
-        <div v-else-if="currentView === 'chat'" key="chat_view" class="h-full">
-          <ChatView
-            v-if="chatActive"
-            :session-id="chatSessionId"
-            :question="chatQuestion"
-            :model-provider="selectedProvider"
-            :model-name="selectedModel"
-            :username="currentUser?.username"
-            @back="backToErrorBank"
-          />
-        </div>
-      </Transition>
-
-      <!-- workspace_review 浮动选择面板 -->
-      <SelectionPanel
-        :visible="currentView === 'workspace_review'"
-        :count="selectedIds.size"
-        export-label="导出错题本"
-        :show-save="true"
-        @export="doExport"
-        @save="doSaveToDb"
-        @clear="deselectAll"
-      />
-
-      <!-- AI 分割任务全局遮罩：置于 main 顶层，仅在录题视图且正在分割时显示 -->
-      <SplitLoading v-if="splitting && (currentView === 'workspace' || currentView === 'workspace_review')" />
-    </main>
-
-    <!-- 全局弹窗与通知 -->
-    <Teleport to="body">
-      <ImageModal
-        :open="modalOpen"
-        :src="modalSrc"
-        :scale="modalScale"
-        @close="closeModal"
-        @update:scale="(s) => modalScale = s"
-      />
-      <ToastContainer :toasts="toasts" />
-
-      <!-- 答案录入弹窗（AI 辅导前置） -->
-      <div v-if="answerModalOpen" class="fixed inset-0 z-[100] flex items-center justify-center p-4">
-        <div class="absolute inset-0 bg-slate-900/40 backdrop-blur-sm dark:bg-black/60" @click="answerModalOpen = false"></div>
-        <div class="relative w-full max-w-lg rounded-2xl border border-slate-200/60 bg-white p-6 shadow-2xl dark:border-white/10 dark:bg-slate-900">
-          <h3 class="mb-1 text-lg font-bold text-slate-900 dark:text-white">录入答案</h3>
-          <p class="mb-4 text-xs text-slate-500 dark:text-slate-400">
-            AI 辅导需要正确答案作为参考。支持 Markdown 格式，数学公式使用 LaTeX（$..$ 行内，$$...$$ 独占行）
-          </p>
-          <textarea
-            v-model="answerModalText"
-            rows="10"
-            placeholder="在此粘贴或输入答案/解析..."
-            class="w-full resize-none rounded-xl border border-slate-200/80 bg-slate-50 px-4 py-3 font-mono text-sm text-slate-800 placeholder-slate-400 focus:border-blue-400 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-white/10 dark:bg-slate-800 dark:text-slate-200"
-          ></textarea>
-          <div class="mt-4 flex justify-end gap-3">
-            <button
-              @click="answerModalOpen = false"
-              class="rounded-xl border border-slate-200/60 bg-white px-5 py-2.5 text-sm font-semibold text-slate-700 transition-colors hover:bg-slate-50 dark:border-white/10 dark:bg-slate-800 dark:text-slate-300 dark:hover:bg-slate-700"
-            >
-              取消
-            </button>
-            <button
-              @click="saveAnswerAndChat"
-              :disabled="answerModalSaving"
-              class="rounded-xl bg-blue-600 px-5 py-2.5 text-sm font-bold text-white shadow-sm transition-all hover:bg-blue-700 disabled:opacity-50 dark:bg-indigo-500 dark:hover:bg-indigo-600"
-            >
-              {{ answerModalSaving ? '保存中...' : '保存并开始辅导' }}
-            </button>
-          </div>
-        </div>
-      </div>
-    </Teleport>
-  </div>
-</template>
-
-<style>
-/* 自定义背景光晕动画 - 极致性能优化版 */
-@keyframes blob {
-  0% { transform: translate3d(0, 0, 0) scale(1); }
-  33% { transform: translate3d(80px, -120px, 0) scale(1.15); }
-  66% { transform: translate3d(-60px, 60px, 0) scale(0.9); }
-  100% { transform: translate3d(0, 0, 0) scale(1); }
-}
-
-.animate-blob {
-  animation: blob 30s infinite alternate ease-in-out;
-  will-change: transform;
-}
-
-.animation-delay-2000 {
-  animation-delay: 2s;
-}
-
-.animation-delay-4000 {
-  animation-delay: 4s;
-}
-
-::view-transition-old(root),
-::view-transition-new(root) {
-  animation: none;
-  mix-blend-mode: normal;
-}
-
-
-.ws-loading-fade-leave-active { transition: opacity 0.4s ease; }
-.ws-loading-fade-leave-to { opacity: 0; }
-
-.ws-loading-bar {
-  animation: wsLoadProgress 2s cubic-bezier(0.4, 0, 0.2, 1) forwards;
-}
-@keyframes wsLoadProgress {
-  from { width: 0%; }
-  to   { width: 100%; }
-}
-
-/* 视图切换：极简淡入淡出 (Simple Fade) */
-.view-fade-enter-active,
-.view-fade-leave-active {
-  transition: opacity 0.3s ease;
-}
-
-.view-fade-enter-from,
-.view-fade-leave-to {
-  opacity: 0;
-}
-
-/* 工作台内部页面切换：左右滑动淡入 (Flip) */
-.flip-enter-active,
-.flip-leave-active {
-  transition: all 0.4s cubic-bezier(0.4, 0, 0.2, 1);
-}
-
-.flip-enter-from {
-  opacity: 0;
-  transform: translateX(20px);
-}
-
-.flip-leave-to {
-  opacity: 0;
-  transform: translateX(-20px);
-}
-
-</style>
diff --git a/frontend/src/views/app/AppLayout.vue b/frontend/src/views/app/AppLayout.vue
new file mode 100644
index 00000000..efdb77e2
--- /dev/null
+++ b/frontend/src/views/app/AppLayout.vue
@@ -0,0 +1,221 @@
+<script setup>
+/**
+ * AppLayout.vue
+ * App 布局容器 — 侧边栏导航 + 内容区视图切换
+ * 仅负责：导航、布局、全局状态（主题/认证/Toast/弹窗）
+ */
+import { computed, nextTick, onBeforeUnmount, onMounted, provide, ref, watch } from 'vue'
+import { useRouter } from 'vue-router'
+import { useAuth } from '@/composables/useAuth.js'
+import { usePageTransition } from '@/composables/usePageTransition.js'
+import { useTheme } from '@/composables/useTheme.js'
+import { useImageModal } from '@/composables/useImageModal.js'
+import { useWorkspaceToast } from '@/composables/useWorkspaceToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import { useSidebarIndicator } from '@/composables/useSidebarIndicator.js'
+import { useAiChatSessions } from '@/composables/useAiChatSessions.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+import { useChatSession } from '@/composables/useChatSession.js'
+import SidebarNav from '@/components/workspace/SidebarNav.vue'
+import ImageModal from '@/components/base/ImageModal.vue'
+import ToastContainer from '@/components/base/ToastContainer.vue'
+import WorkspaceBackground from '@/components/workspace/WorkspaceBackground.vue'
+import AnswerInputModal from '@/components/workspace/AnswerInputModal.vue'
+import WorkspaceView from '@/views/app/WorkspaceView.vue'
+import Dashboard from '@/views/app/DashboardView.vue'
+import ReviewView from '@/views/app/ReviewView.vue'
+import ErrorBank from '@/views/app/ErrorBankView.vue'
+import ChatView from '@/views/app/ChatView.vue'
+import SettingsView from '@/views/app/SettingsView.vue'
+import NoteView from '@/views/app/NoteView.vue'
+import ChatPage from '@/views/app/ChatPageView.vue'
+
+// ── 全局 Composables ────────────────────────────────────
+const router = useRouter()
+const { currentUser } = useAuth()
+const { loading: globalLoading } = usePageTransition()
+const { isDark, toggleTheme, initTheme } = useTheme()
+const { toasts, pushToast } = useWorkspaceToast()
+const { modalOpen, modalSrc, modalScale, closeModal } = useImageModal()
+const { doFetchStatus } = useSystemStatus()
+const {
+  navRef, navBtnRefs, indicatorStyle, indicatorTransition,
+  chatListRef, chatBtnRefs, chatIndicatorStyle, chatIndicatorTransition,
+  updateIndicator: _updateIndicator,
+} = useSidebarIndicator()
+const {
+  aiChatSessions, activeAiChatId,
+  chatMenuOpenId, renamingChatId, renameText,
+  loadAiChatSessions, createAiChat: _createAiChat, selectAiChat: _selectAiChat,
+  toggleChatMenu, closeChatMenu,
+  startRenameChat, confirmRenameChat, deleteAiChat,
+} = useAiChatSessions(pushToast)
+const {
+  currentView, lastWorkspaceView, collapsedGroups, chatCollapsed,
+  NAV_GROUPS, WORKSPACE_VIEWS, navigateToHome,
+} = useWorkspaceNav()
+const {
+  answerModalOpen, answerModalText, answerModalSaving,
+  saveAnswerAndChat,
+} = useChatSession()
+
+provide('pushToast', pushToast) // useToast() composable injects this key
+
+// ── 认证 ────────────────────────────────────────────────
+const theme = computed(() => isDark.value ? 'dark' : 'light')
+const userMenuOpen = ref(false)
+
+const handleLogout = async () => {
+  try { await fetch('/api/auth/logout', { method: 'POST' }) } catch (_) {}
+  currentUser.value = null
+  router.push('/auth')
+}
+
+// ── 指示器更新 ──────────────────────────────────────────
+const updateIndicator = (animate = true) => {
+  _updateIndicator(currentView.value, activeAiChatId.value, NAV_GROUPS, collapsedGroups.value, animate)
+}
+
+watch(currentView, () => nextTick(updateIndicator))
+watch(collapsedGroups, () => {
+  updateIndicator(false)
+  nextTick(() => updateIndicator(false))
+}, { deep: true })
+
+// ── AI 独立对话包装 ────────────────────────────────────
+const createAiChat = () => _createAiChat(currentView)
+const selectAiChat = (s) => _selectAiChat(s, currentView)
+
+// ── 键盘事件 ────────────────────────────────────────────
+const onKeydown = (e) => {
+  if (e.key === 'Escape' && modalOpen.value) closeModal()
+}
+
+// ── 视图切换 ────────────────────────────────────────────
+watch(currentView, (newView) => {
+  if (newView === 'ai-chat') loadAiChatSessions()
+})
+
+// ── 生命周期 ────────────────────────────────────────────
+onMounted(() => {
+  initTheme()
+  document.addEventListener('keydown', onKeydown)
+  document.addEventListener('click', closeChatMenu)
+  loadAiChatSessions()
+  nextTick(updateIndicator)
+  watch(() => activeAiChatId.value, () => nextTick(updateIndicator))
+
+  if (!globalLoading.value) {
+    doFetchStatus()
+  } else {
+    const unwatch = watch(globalLoading, (val) => {
+      if (!val) { doFetchStatus(); unwatch() }
+    })
+  }
+})
+
+onBeforeUnmount(() => {
+  document.removeEventListener('keydown', onKeydown)
+  document.removeEventListener('click', closeChatMenu)
+})
+</script>
+
+<template>
+  <div class="flex h-screen w-full overflow-hidden bg-[#0c0c0e] font-sans text-slate-300 relative">
+
+    <WorkspaceBackground />
+
+    <!-- 侧边栏导航 -->
+    <SidebarNav
+      :current-view="currentView"
+      :last-workspace-view="lastWorkspaceView"
+      :current-user="currentUser"
+      :is-dark="isDark"
+      :theme="theme"
+      :nav-groups="NAV_GROUPS"
+      :workspace-views="WORKSPACE_VIEWS"
+      :collapsed-groups="collapsedGroups"
+      :nav-btn-refs="navBtnRefs"
+      :indicator-style="indicatorStyle"
+      :indicator-transition="indicatorTransition"
+      :chat-collapsed="chatCollapsed"
+      :ai-chat-sessions="aiChatSessions"
+      :active-ai-chat-id="activeAiChatId"
+      :chat-btn-refs="chatBtnRefs"
+      :chat-indicator-style="chatIndicatorStyle"
+      :chat-indicator-transition="chatIndicatorTransition"
+      :chat-menu-open-id="chatMenuOpenId"
+      :renaming-chat-id="renamingChatId"
+      :rename-text="renameText"
+      :user-menu-open="userMenuOpen"
+      @update:current-view="(v) => currentView = v"
+      @update:collapsed-groups="(v) => Object.assign(collapsedGroups, v)"
+      @update:chat-collapsed="(v) => chatCollapsed = v"
+      @update:user-menu-open="(v) => userMenuOpen = v"
+      @update:chat-menu-open-id="(v) => chatMenuOpenId = v"
+      @update:rename-text="(v) => renameText = v"
+      @update:nav-ref="(el) => navRef = el"
+      @navigate-home="navigateToHome"
+      @logout="handleLogout"
+      @toggle-theme="toggleTheme"
+      @create-ai-chat="createAiChat"
+      @select-ai-chat="selectAiChat"
+      @start-rename-chat="startRenameChat"
+      @confirm-rename-chat="confirmRenameChat"
+      @delete-ai-chat="deleteAiChat"
+      @toggle-chat-menu="toggleChatMenu"
+    />
+
+    <!-- ================== 右侧内容区 ================== -->
+    <div class="relative z-10 flex-1 overflow-hidden pb-20 md:pb-3 md:pt-3 md:pr-3">
+      <Transition name="view-fade" mode="out-in">
+        <WorkspaceView v-if="currentView === 'workspace' || currentView === 'workspace_review'" key="workspace" :theme="theme" />
+        <ReviewView v-else-if="currentView === 'review'" key="review" />
+        <Dashboard v-else-if="currentView === 'dashboard'" key="dashboard" />
+        <ErrorBank v-else-if="currentView === 'error-bank'" key="error-bank" />
+        <SettingsView v-else-if="currentView === 'settings'" key="settings" />
+        <ChatView v-else-if="currentView === 'chat'" key="chat" />
+        <NoteView v-else-if="currentView === 'notes'" key="notes" />
+        <ChatPage v-else-if="currentView === 'ai-chat'" key="ai-chat" />
+      </Transition>
+    </div>
+
+    <!-- 全局弹窗与通知 -->
+    <Teleport to="body">
+      <ImageModal
+        :open="modalOpen"
+        :src="modalSrc"
+        :scale="modalScale"
+        @close="closeModal"
+        @update:scale="(s) => modalScale = s"
+      />
+      <ToastContainer :toasts="toasts" />
+      <AnswerInputModal
+        :open="answerModalOpen"
+        :text="answerModalText"
+        :saving="answerModalSaving"
+        @update:open="(v) => answerModalOpen = v"
+        @update:text="(v) => answerModalText = v"
+        @confirm="saveAnswerAndChat"
+      />
+    </Teleport>
+  </div>
+</template>
+
+<style>
+::view-transition-old(root),
+::view-transition-new(root) {
+  animation: none;
+  mix-blend-mode: normal;
+}
+
+/* 视图切换淡入淡出 */
+.view-fade-enter-active,
+.view-fade-leave-active {
+  transition: opacity 0.3s ease;
+}
+.view-fade-enter-from,
+.view-fade-leave-to {
+  opacity: 0;
+}
+</style>
diff --git a/frontend/src/views/app/ChatPageView.vue b/frontend/src/views/app/ChatPageView.vue
new file mode 100644
index 00000000..33dc103a
--- /dev/null
+++ b/frontend/src/views/app/ChatPageView.vue
@@ -0,0 +1,286 @@
+<script setup>
+import { ref, watch, nextTick, computed } from 'vue'
+import * as api from '@/api.js'
+import { renderMarkdown, typesetMath } from '@/utils.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import { useToast } from '@/composables/useToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import { useAuth } from '@/composables/useAuth.js'
+import { useAiChatSessions } from '@/composables/useAiChatSessions.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+
+const { pushToast } = useToast()
+const { selectedProvider, selectedModel } = useSystemStatus()
+const { currentUser } = useAuth()
+const { activeAiChatId, createAiChat, onAiChatTitleUpdated } = useAiChatSessions(pushToast)
+const { currentView } = useWorkspaceNav()
+
+const sessionId = computed(() => activeAiChatId.value)
+const modelProvider = computed(() => selectedProvider.value)
+const modelName = computed(() => selectedModel.value)
+const username = computed(() => currentUser.value?.username || '')
+
+// ---- 对话消息 ----
+const messages = ref([])
+const inputText = ref('')
+const streaming = ref(false)
+const messagesContainer = ref(null)
+const deepThink = ref(false)
+
+watch(sessionId, (id) => {
+  if (id) loadMessages()
+  else messages.value = []
+}, { immediate: true })
+
+async function loadMessages() {
+  if (!sessionId.value) return
+  try {
+    const result = await api.fetchMessages(sessionId.value, { limit: 50 })
+    messages.value = result.messages || []
+    await nextTick()
+    scrollToBottom()
+    if (messagesContainer.value) typesetMath(messagesContainer.value)
+  } catch (e) {
+    pushToast('error', e.message)
+  }
+}
+
+function scrollToBottom() {
+  if (messagesContainer.value) {
+    messagesContainer.value.scrollTop = messagesContainer.value.scrollHeight
+  }
+}
+
+async function sendMessage() {
+  const text = inputText.value.trim()
+  if (!text || !sessionId.value || streaming.value) return
+
+  inputText.value = ''
+  // 消息结构：{ role, content, reasoning?, reasoningOpen? }
+  messages.value.push({ role: 'user', content: text })
+  messages.value.push({ role: 'assistant', content: '', reasoning: '', reasoningOpen: false })
+  await nextTick()
+  scrollToBottom()
+  // 重置 textarea 高度
+  if (textareaRef.value) textareaRef.value.style.height = 'auto'
+
+  streaming.value = true
+  const useDeepThink = deepThink.value
+
+  try {
+    const resp = await api.streamChat(
+      sessionId.value,
+      text,
+      modelProvider.value,
+      null,
+      modelName.value,
+      { deepThink: useDeepThink },
+    )
+
+    const reader = resp.body.getReader()
+    const decoder = new TextDecoder()
+    let buffer = ''
+    const lastMsg = () => messages.value[messages.value.length - 1]
+
+    while (true) {
+      const { done, value } = await reader.read()
+      if (done) break
+      buffer += decoder.decode(value, { stream: true })
+
+      const lines = buffer.split('\n')
+      buffer = lines.pop() || ''
+
+      for (const line of lines) {
+        if (!line.startsWith('data: ')) continue
+        try {
+          const payload = JSON.parse(line.slice(6))
+          if (payload.reasoning) {
+            lastMsg().reasoning += payload.reasoning
+            lastMsg().reasoningOpen = true  // 思考中自动展开
+            await nextTick()
+            scrollToBottom()
+          }
+          if (payload.token) {
+            // 收到正文 token 时折叠思考过程
+            if (lastMsg().reasoningOpen && lastMsg().reasoning) {
+              lastMsg().reasoningOpen = false
+            }
+            lastMsg().content += payload.token
+            await nextTick()
+            scrollToBottom()
+          }
+          if (payload.done) {
+            const firstMsg = text.slice(0, 20) + (text.length > 20 ? '...' : '')
+            onAiChatTitleUpdated(sessionId.value, firstMsg)
+          }
+          if (payload.error) {
+            lastMsg().content += `\n\n⚠️ ${payload.error}`
+          }
+        } catch (_) {}
+      }
+    }
+
+    await nextTick()
+    if (messagesContainer.value) typesetMath(messagesContainer.value)
+  } catch (e) {
+    if (e.name !== 'AbortError') {
+      messages.value[messages.value.length - 1].content += `\n\n⚠️ ${e.message}`
+    }
+  } finally {
+    streaming.value = false
+  }
+}
+
+const textareaRef = ref(null)
+
+function handleKeydown(e) {
+  if (e.key === 'Enter' && !e.shiftKey) {
+    e.preventDefault()
+    sendMessage()
+  }
+}
+
+function autoResize() {
+  const el = textareaRef.value
+  if (!el) return
+  el.style.height = 'auto'
+  el.style.height = Math.min(el.scrollHeight, 200) + 'px'
+}
+</script>
+
+<template>
+  <ContentPanel title="AI 对话">
+  <div class="flex flex-col h-full">
+    <!-- 消息区域（含空状态） -->
+    <div ref="messagesContainer" class="flex-1 overflow-y-auto custom-scrollbar">
+      <div class="mx-auto max-w-5xl px-4 sm:px-8">
+
+        <!-- 空状态：居中问候 -->
+        <div v-if="messages.length === 0 && !streaming" class="flex flex-col items-center justify-center" style="min-height: calc(100vh - 200px)">
+          <p class="text-2xl font-bold text-[#f7f8f8]">
+            Hi，{{ username || '同学' }}
+          </p>
+          <p class="mt-2 text-sm text-[#62666d]">有问题，尽管问</p>
+        </div>
+
+        <!-- 消息列表 -->
+        <div v-else class="space-y-6 py-6">
+          <div v-for="(msg, i) in messages" :key="i" class="flex" :class="msg.role === 'user' ? 'justify-end' : 'justify-start'">
+            <div
+              class="max-w-[85%] rounded-2xl px-4 py-3 text-sm leading-relaxed"
+              :class="msg.role === 'user'
+                ? 'brand-btn text-white rounded-br-lg'
+                : 'brand-btn text-[#d0d6e0] rounded-bl-lg'"
+            >
+              <!-- 思考过程折叠面板 -->
+              <div v-if="msg.role === 'assistant' && msg.reasoning" class="mb-3">
+                <button
+                  @click="msg.reasoningOpen = !msg.reasoningOpen"
+                  class="flex items-center gap-2 text-xs font-bold text-[#8a8f98] hover:text-[#d0d6e0] transition-colors"
+                >
+                  <i class="fa-solid fa-brain text-[rgb(129,115,223)]"></i>
+                  <span>{{ streaming && i === messages.length - 1 && !msg.content ? '思考中...' : '已深度思考' }}</span>
+                  <i class="fa-solid text-[10px] transition-transform" :class="msg.reasoningOpen ? 'fa-chevron-up' : 'fa-chevron-down'"></i>
+                </button>
+                <Transition
+                  enter-active-class="transition-all duration-200 ease-out"
+                  enter-from-class="opacity-0 max-h-0"
+                  enter-to-class="opacity-100 max-h-[400px]"
+                  leave-active-class="transition-all duration-150 ease-in"
+                  leave-from-class="opacity-100 max-h-[400px]"
+                  leave-to-class="opacity-0 max-h-0"
+                >
+                  <div v-if="msg.reasoningOpen" class="mt-2 overflow-auto max-h-[400px] rounded-xl bg-white/[0.03] p-3 text-xs text-[#8a8f98] leading-relaxed custom-scrollbar whitespace-pre-wrap">{{ msg.reasoning }}</div>
+                </Transition>
+              </div>
+              <!-- 正文内容 -->
+              <div v-if="msg.role === 'assistant'" v-html="renderMarkdown(msg.content)" class="prose prose-sm prose-invert max-w-none"></div>
+              <div v-else class="whitespace-pre-wrap">{{ msg.content }}</div>
+            </div>
+          </div>
+
+          <!-- 流式加载指示 -->
+          <div v-if="streaming && messages.length && !messages[messages.length - 1].content" class="flex justify-start">
+            <div class="brand-btn rounded-2xl rounded-bl-lg px-4 py-3">
+              <div class="flex gap-1">
+                <span class="w-2 h-2 rounded-full bg-[#62666d] animate-bounce" style="animation-delay: 0ms"></span>
+                <span class="w-2 h-2 rounded-full bg-[#62666d] animate-bounce" style="animation-delay: 150ms"></span>
+                <span class="w-2 h-2 rounded-full bg-[#62666d] animate-bounce" style="animation-delay: 300ms"></span>
+              </div>
+            </div>
+          </div>
+        </div>
+
+      </div>
+    </div>
+
+    <!-- 底部输入区域 -->
+    <div class="shrink-0 px-4 pb-4 pt-2 sm:px-8">
+      <div class="mx-auto max-w-5xl">
+        <div class="rounded-xl brand-btn overflow-hidden">
+          <!-- 文本输入 -->
+          <textarea
+            ref="textareaRef"
+            v-model="inputText"
+            @keydown="handleKeydown"
+            @input="autoResize"
+            :disabled="streaming || !sessionId"
+            rows="1"
+            :placeholder="sessionId ? '有问题，尽管问，shift+enter 换行' : '请先新建或选择一个对话'"
+            class="w-full resize-none bg-transparent px-4 pt-3 pb-1 text-sm outline-none text-[#f7f8f8] placeholder-[#62666d]"
+            style="min-height: 24px; max-height: 200px;"
+          ></textarea>
+
+          <!-- 底部工具栏 -->
+          <div class="flex items-center justify-between px-3 pb-2.5 pt-1">
+            <!-- 左侧功能按钮 -->
+            <div class="flex items-center gap-0.5">
+              <button
+                @click="deepThink = !deepThink"
+                class="h-8 px-2.5 rounded-lg text-xs font-bold flex items-center gap-1.5 transition-colors"
+                :class="deepThink
+                  ? 'bg-[rgb(129,115,223)]/15 text-[rgb(145,132,235)]'
+                  : 'text-[#62666d] hover:text-[#8a8f98]'"
+                title="深度思考"
+              >
+                <i class="fa-solid fa-brain text-sm"></i>
+                <span class="hidden sm:inline">深度思考</span>
+              </button>
+              <button
+                disabled
+                class="h-8 px-2.5 rounded-lg text-xs font-bold flex items-center gap-1.5 text-[#62666d] cursor-not-allowed"
+                title="联网搜索（敬请期待）"
+              >
+                <i class="fa-solid fa-globe text-sm"></i>
+                <span class="hidden sm:inline">联网搜索</span>
+              </button>
+            </div>
+
+            <!-- 右侧操作按钮 -->
+            <div class="flex items-center gap-1.5">
+              <button
+                disabled
+                class="h-8 w-8 rounded-lg flex items-center justify-center text-[#62666d] hover:bg-white/[0.04] transition-colors cursor-not-allowed"
+                title="附件（敬请期待）"
+              >
+                <i class="fa-solid fa-plus text-sm"></i>
+              </button>
+              <button
+                @click="sessionId ? sendMessage() : createAiChat(currentView)"
+                :disabled="sessionId ? (!inputText.trim() || streaming) : false"
+                class="h-8 w-8 rounded-full flex items-center justify-center transition-all"
+                :class="inputText.trim() && sessionId
+                  ? 'brand-btn text-white'
+                  : 'bg-white/[0.06] text-[#62666d] opacity-50'"
+              >
+                <i class="fa-solid fa-arrow-up text-xs"></i>
+              </button>
+            </div>
+          </div>
+        </div>
+        <p class="mt-2 text-center text-xs text-[#62666d]">内容由 AI 生成，仅供参考</p>
+      </div>
+    </div>
+  </div>
+  </ContentPanel>
+</template>
diff --git a/frontend/src/components/ChatView.vue b/frontend/src/views/app/ChatView.vue
similarity index 87%
rename from frontend/src/components/ChatView.vue
rename to frontend/src/views/app/ChatView.vue
index a10833f1..1bfab67b 100644
--- a/frontend/src/components/ChatView.vue
+++ b/frontend/src/views/app/ChatView.vue
@@ -1,21 +1,25 @@
 <script setup>
-import { ref, computed, nextTick, onMounted, onBeforeUnmount, watch, inject } from 'vue'
-import { fetchMessages, streamChat } from '../api.js'
-import { getQuestionSnippet, renderMarkdown, typesetMath } from '../utils.js'
+import { ref, computed, nextTick, onMounted, onBeforeUnmount, watch } from 'vue'
+import { fetchMessages, streamChat } from '@/api.js'
+import { getQuestionSnippet, renderMarkdown, typesetMath } from '@/utils.js'
+import { useToast } from '@/composables/useToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import { useAuth } from '@/composables/useAuth.js'
+import { useChatSession } from '@/composables/useChatSession.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
 
 const PAGE_SIZE = 30
 
-const props = defineProps({
-  sessionId: { type: Number, default: null },
-  question: { type: Object, default: null },
-  modelProvider: { type: String, default: 'openai' },
-  modelName: { type: String, default: '' },
-  username: { type: String, default: '' },
-})
-
-const emit = defineEmits(['back'])
+const { pushToast } = useToast()
+const { selectedProvider, selectedModel } = useSystemStatus()
+const { currentUser } = useAuth()
+const { chatSessionId, chatQuestion, backToErrorBank } = useChatSession()
 
-const pushToast = inject('pushToast', (type, msg) => { console.warn(`[${type}] ${msg}`) })
+const sessionId = computed(() => chatSessionId.value)
+const question = computed(() => chatQuestion.value)
+const modelProvider = computed(() => selectedProvider.value)
+const modelName = computed(() => selectedModel.value)
+const username = computed(() => currentUser.value?.username || '')
 
 const toMsg = (m) => ({ id: m.id, role: m.role, content: m.content })
 
@@ -27,7 +31,7 @@ const snippetEl = ref(null)
 const hasMore = ref(false)
 const loadingMore = ref(false)
 
-const snippet = computed(() => getQuestionSnippet(props.question, 0, '未知题目'))
+const snippet = computed(() => getQuestionSnippet(question.value, 0, '未知题目'))
 
 let abortCtrl = null
 let scrollRafId = null
@@ -46,9 +50,9 @@ const scrollToBottom = async () => {
 }
 
 const loadHistory = async () => {
-  if (!props.sessionId) return
+  if (!sessionId.value) return
   try {
-    const data = await fetchMessages(props.sessionId, { limit: PAGE_SIZE })
+    const data = await fetchMessages(sessionId.value, { limit: PAGE_SIZE })
     messages.value = data.messages.map(toMsg)
     hasMore.value = data.hasMore
     scrollToBottom()
@@ -60,7 +64,7 @@ const loadHistory = async () => {
 }
 
 const loadOlder = async () => {
-  if (!hasMore.value || loadingMore.value || !props.sessionId) return
+  if (!hasMore.value || loadingMore.value || !sessionId.value) return
   const firstMsg = messages.value[0]
   if (!firstMsg || !firstMsg.id) return
 
@@ -70,7 +74,7 @@ const loadOlder = async () => {
     const el = listEl.value
     const prevHeight = el ? el.scrollHeight : 0
 
-    const data = await fetchMessages(props.sessionId, {
+    const data = await fetchMessages(sessionId.value, {
       limit: PAGE_SIZE,
       beforeId: firstMsg.id,
     })
@@ -104,7 +108,7 @@ const onScroll = () => {
 
 const sendMessage = async () => {
   const text = inputText.value.trim()
-  if (!text || streaming.value || !props.sessionId) return
+  if (!text || streaming.value || !sessionId.value) return
 
   inputText.value = ''
   messages.value.push({ role: 'user', content: text })
@@ -114,7 +118,7 @@ const sendMessage = async () => {
 
   abortCtrl = new AbortController()
   try {
-    const resp = await streamChat(props.sessionId, text, props.modelProvider, abortCtrl.signal, props.modelName)
+    const resp = await streamChat(sessionId.value, text, modelProvider.value, abortCtrl.signal, modelName.value)
     if (!resp.ok) {
       const err = await resp.json().catch(() => null)
       throw new Error((err && err.error) || `HTTP ${resp.status}`)
@@ -170,7 +174,7 @@ onMounted(async () => {
   await nextTick()
   typesetMath(snippetEl.value)
 })
-watch(() => props.sessionId, () => {
+watch(sessionId, () => {
   abortCtrl?.abort()
   mdCache.clear()
   loadHistory()
@@ -187,13 +191,14 @@ onBeforeUnmount(() => {
 </script>
 
 <template>
+  <ContentPanel title="AI 辅导">
   <div class="flex h-full flex-col">
     <!-- 顶栏 -->
     <div
-      class="flex shrink-0 items-center gap-3 border-b border-slate-200/60 bg-white/70 px-5 py-4 backdrop-blur-xl dark:border-white/10 dark:bg-[#0A0A0F]/60"
+      class="flex shrink-0 items-center gap-3 border-b border-slate-200/60 bg-white/70 px-5 py-4 dark:border-white/10 dark:bg-[#0A0A0F]/60"
     >
       <button
-        @click="emit('back')"
+        @click="backToErrorBank()"
         class="flex h-9 w-9 items-center justify-center rounded-lg text-slate-500 transition-colors hover:bg-slate-100 hover:text-slate-700 dark:text-slate-400 dark:hover:bg-white/5 dark:hover:text-white"
       >
         <i class="fa-solid fa-arrow-left"></i>
@@ -306,7 +311,7 @@ onBeforeUnmount(() => {
 
     <!-- 输入区域 -->
     <div
-      class="shrink-0 border-t border-slate-200/60 bg-white/70 px-4 py-4 backdrop-blur-xl dark:border-white/10 dark:bg-[#0A0A0F]/60 sm:px-6"
+      class="shrink-0 border-t border-slate-200/60 bg-white/70 px-4 py-4 dark:border-white/10 dark:bg-[#0A0A0F]/60 sm:px-6"
     >
       <div class="mx-auto flex max-w-3xl items-end gap-3">
         <textarea
@@ -327,4 +332,5 @@ onBeforeUnmount(() => {
       </div>
     </div>
   </div>
+  </ContentPanel>
 </template>
diff --git a/frontend/src/components/Dashboard.vue b/frontend/src/views/app/DashboardView.vue
similarity index 82%
rename from frontend/src/components/Dashboard.vue
rename to frontend/src/views/app/DashboardView.vue
index f8e68e46..844cc948 100644
--- a/frontend/src/components/Dashboard.vue
+++ b/frontend/src/views/app/DashboardView.vue
@@ -1,17 +1,18 @@
 <script setup>
-import { ref, watch, nextTick, onBeforeUnmount, computed } from 'vue'
-import * as api from '../api.js'
-import CustomSelect from './CustomSelect.vue'
-import GlassCard from './GlassCard.vue'
-import PageHeader from './PageHeader.vue'
-import StatCard from './StatCard.vue'
+import { ref, watch, nextTick, onMounted, onBeforeUnmount, computed } from 'vue'
+import * as api from '@/api.js'
+import { useTheme } from '@/composables/useTheme.js'
+import { useToast } from '@/composables/useToast.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import BaseSelect from '@/components/base/BaseSelect.vue'
+import BaseCard from '@/components/base/BaseCard.vue'
+import StatCard from '@/components/dashboard/StatCard.vue'
 
-const props = defineProps({
-  theme: { type: String, default: 'light' },
-  visible: { type: Boolean, default: false },
-})
-
-const emit = defineEmits(['go-workspace', 'push-toast'])
+const { isDark } = useTheme()
+const { pushToast } = useToast()
+const { currentView } = useWorkspaceNav()
+const theme = computed(() => isDark.value ? 'dark' : 'light')
 
 // ---- 统计数据 ----
 const stats = ref(null)
@@ -40,7 +41,7 @@ const loadStats = async () => {
     const data = await api.fetchDashboardStats(selectedSubject.value || undefined)
     stats.value = data
   } catch (e) {
-    emit('push-toast', 'error', '加载统计数据失败')
+    pushToast('error', '加载统计数据失败')
   } finally {
     statsLoading.value = false
   }
@@ -53,9 +54,9 @@ watch(selectedSubject, () => { loadStats() })
 const initCharts = async () => {
   await nextTick()
   if (!window.Chart || !stats.value) return
-  const isDark = props.theme === 'dark'
-  const textColor = isDark ? 'rgba(255,255,255,0.7)' : 'rgba(15,23,42,0.8)'
-  const gridColor = isDark ? 'rgba(255,255,255,0.05)' : 'rgba(15,23,42,0.08)'
+  const dark = isDark.value
+  const textColor = dark ? 'rgba(255,255,255,0.7)' : 'rgba(15,23,42,0.8)'
+  const gridColor = dark ? 'rgba(255,255,255,0.05)' : 'rgba(15,23,42,0.08)'
 
   initTrendChart(isDark, textColor, gridColor)
   initBarChart(isDark, textColor, gridColor)
@@ -173,9 +174,9 @@ const heatmapMax = computed(() => {
 })
 
 const heatmapCellColor = (val) => {
-  if (!val) return props.theme === 'dark' ? 'rgba(255,255,255,0.03)' : 'rgba(15,23,42,0.03)'
+  if (!val) return theme.value === 'dark' ? 'rgba(255,255,255,0.03)' : 'rgba(15,23,42,0.03)'
   const ratio = val / heatmapMax.value
-  if (props.theme === 'dark') {
+  if (theme.value === 'dark') {
     const alpha = 0.15 + ratio * 0.7
     return `rgba(99, 102, 241, ${alpha})`
   }
@@ -184,12 +185,10 @@ const heatmapCellColor = (val) => {
 }
 
 // ---- 生命周期 ----
-watch(() => props.visible, (v) => {
-  if (v) loadStats()
-}, { immediate: true })
+onMounted(() => loadStats())
 
-watch(() => [props.theme, stats.value], () => {
-  if (props.visible && stats.value) nextTick(initCharts)
+watch(() => [isDark.value, stats.value], () => {
+  if (stats.value) nextTick(initCharts)
 })
 
 onBeforeUnmount(() => {
@@ -200,29 +199,22 @@ onBeforeUnmount(() => {
 </script>
 
 <template>
+  <ContentPanel title="数据面板">
+    <template #toolbar>
+      <BaseSelect v-if="subjects.length" v-model="selectedSubject" :options="subjects" placeholder="全部学科" width-class="min-w-[140px]" />
+      <button @click="currentView = 'workspace'" class="inline-flex items-center gap-2 rounded-md brand-btn px-3 py-1.5 text-xs font-medium text-[#f7f8f8]">
+        <i class="fa-solid fa-plus-circle text-[10px]"></i> 录入新题目
+      </button>
+    </template>
   <div class="relative h-full overflow-y-auto custom-scrollbar">
     <div class="container relative z-10 mx-auto max-w-6xl px-4 py-8 sm:px-8">
-      <!-- 页面标题 + 学科筛选 -->
-      <PageHeader
-        title="数据面板"
-        subtitle="错题统计与学习趋势一览"
-      >
-        <template #extra>
-          <div class="flex items-center gap-4">
-            <CustomSelect v-if="subjects.length" v-model="selectedSubject" :options="subjects" placeholder="全部学科" width-class="min-w-[140px]" />
-            <button @click="emit('go-workspace')" class="btn-primary group h-10 px-8 shadow-md shadow-blue-500/20">
-              <i class="fa-solid fa-plus-circle transition-transform group-hover:rotate-90"></i> 录入新题目
-            </button>
-          </div>
-        </template>
-      </PageHeader>
 
       <!-- 骨架屏（加载中） -->
       <template v-if="statsLoading">
         <!-- 统计卡片骨架：精确复刻 StatCard 内部结构 -->
         <div class="mb-8 grid grid-cols-2 gap-4 sm:grid-cols-4">
           <div v-for="i in 4" :key="i"
-            class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]">
+            class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm dark:border-white/10 dark:bg-white/[0.03]">
             <div class="flex items-center gap-4">
               <div class="size-12 shrink-0 rounded-lg bg-slate-200/80 dark:bg-white/[0.07]"></div>
               <div class="flex flex-col gap-2">
@@ -232,10 +224,10 @@ onBeforeUnmount(() => {
             </div>
           </div>
         </div>
-        <!-- 图表骨架：精确复刻 GlassCard + 标题行 + canvas 区域 -->
+        <!-- 图表骨架：精确复刻 BaseCard + 标题行 + canvas 区域 -->
         <div class="mb-6 grid grid-cols-1 gap-6 lg:grid-cols-2">
           <div v-for="i in 2" :key="i"
-            class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]">
+            class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm dark:border-white/10 dark:bg-white/[0.03]">
             <div class="mb-4 flex items-center gap-2">
               <div class="h-4 w-4 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
               <div class="h-4 w-28 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
@@ -244,14 +236,14 @@ onBeforeUnmount(() => {
           </div>
         </div>
         <div class="mb-8 grid grid-cols-1 gap-6 lg:grid-cols-2">
-          <div class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]">
+          <div class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm dark:border-white/10 dark:bg-white/[0.03]">
             <div class="mb-4 flex items-center gap-2">
               <div class="h-4 w-4 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
               <div class="h-4 w-28 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
             </div>
             <div class="h-[360px] w-full rounded-xl bg-slate-100/80 dark:bg-white/[0.04]"></div>
           </div>
-          <div class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl dark:border-white/10 dark:bg-white/[0.03]">
+          <div class="animate-pulse rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm dark:border-white/10 dark:bg-white/[0.03]">
             <div class="mb-4 flex items-center gap-2">
               <div class="h-4 w-4 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
               <div class="h-4 w-28 rounded bg-slate-200/80 dark:bg-white/[0.07]"></div>
@@ -268,36 +260,36 @@ onBeforeUnmount(() => {
           <StatCard label="总错题" icon="fa-solid fa-layer-group" :value="stats?.total_questions || 0" unit="道" color="indigo" />
           <StatCard label="待复习" icon="fa-solid fa-clock" :value="stats?.review_stats?.['待复习'] || 0" unit="道" color="blue" />
           <StatCard label="已掌握" icon="fa-solid fa-circle-check" :value="stats?.review_stats?.['已掌握'] || 0" unit="道" color="emerald" />
-          <StatCard label="今日掌握" icon="fa-solid fa-bolt" :value="stats?.review_stats?.['已掌握'] || 0" unit="道" color="slate" />
+          <StatCard label="今日掌握" icon="fa-solid fa-bolt" :value="stats?.today_mastered || 0" unit="道" color="slate" />
         </div>
 
         <!-- 图表区第一行：趋势 + 条形图 -->
         <div class="mb-6 grid grid-cols-1 gap-6 lg:grid-cols-2">
-          <GlassCard>
+          <BaseCard>
             <h3 class="mb-4 flex items-center gap-2 text-sm font-black text-slate-700 dark:text-slate-300">
               <i class="fa-solid fa-chart-line text-blue-500"></i> 最近 30 天趋势
             </h3>
             <div class="relative h-[240px] w-full"><canvas ref="trendCanvas"></canvas></div>
-          </GlassCard>
-          <GlassCard>
+          </BaseCard>
+          <BaseCard>
             <h3 class="mb-4 flex items-center gap-2 text-sm font-black text-slate-700 dark:text-slate-300">
               <i class="fa-solid fa-ranking-star text-indigo-500"></i> Top 10 易错知识点
             </h3>
             <div v-if="stats?.tag_stats?.length" class="relative h-[240px] w-full"><canvas ref="barCanvas"></canvas></div>
             <div v-else class="flex h-[240px] items-center justify-center text-sm text-slate-400">暂无标签数据</div>
-          </GlassCard>
+          </BaseCard>
         </div>
 
         <!-- 图表区第二行：堆叠柱状图 + 热力图 -->
         <div class="mb-8 grid grid-cols-1 gap-6 lg:grid-cols-2">
-          <GlassCard>
+          <BaseCard>
             <h3 class="mb-4 flex items-center gap-2 text-sm font-black text-slate-700 dark:text-slate-300">
               <i class="fa-solid fa-chart-bar text-emerald-500"></i> 知识点掌握状态
             </h3>
             <div v-if="stats?.tag_status_stats?.length" class="relative h-[360px] w-full"><canvas ref="stackedCanvas"></canvas></div>
             <div v-else class="flex h-[360px] items-center justify-center text-sm text-slate-400">暂无掌握状态数据</div>
-          </GlassCard>
-          <GlassCard>
+          </BaseCard>
+          <BaseCard>
             <h3 class="mb-4 flex items-center gap-2 text-sm font-black text-slate-700 dark:text-slate-300">
               <i class="fa-solid fa-fire text-rose-500"></i> 知识点 × 题型分布
             </h3>
@@ -305,13 +297,13 @@ onBeforeUnmount(() => {
               <table class="heatmap-table w-full text-center text-xs">
                 <thead>
                   <tr>
-                    <th class="sticky left-0 z-10 bg-white/80 px-2 py-2 text-left font-bold text-slate-500 backdrop-blur-sm dark:bg-white/[0.03] dark:text-slate-400"></th>
+                    <th class="sticky left-0 z-10 bg-white/80 px-2 py-2 text-left font-bold text-slate-500 dark:bg-white/[0.03] dark:text-slate-400"></th>
                     <th v-for="t in heatmapData.types" :key="t" class="px-2 py-2 font-bold text-slate-500 dark:text-slate-400">{{ t }}</th>
                   </tr>
                 </thead>
                 <tbody>
                   <tr v-for="(tag, ri) in heatmapData.tags" :key="tag">
-                    <td class="sticky left-0 z-10 max-w-[120px] truncate bg-white/80 px-2 py-1.5 text-left font-bold text-slate-600 backdrop-blur-sm dark:bg-white/[0.03] dark:text-slate-300">{{ tag }}</td>
+                    <td class="sticky left-0 z-10 max-w-[120px] truncate bg-white/80 px-2 py-1.5 text-left font-bold text-slate-600 dark:bg-white/[0.03] dark:text-slate-300">{{ tag }}</td>
                     <td v-for="(t, ci) in heatmapData.types" :key="t" class="px-1 py-1">
                       <div
                         class="mx-auto flex h-8 w-full min-w-[40px] items-center justify-center rounded-lg font-bold"
@@ -324,11 +316,12 @@ onBeforeUnmount(() => {
               </table>
             </div>
             <div v-else class="flex h-[240px] items-center justify-center text-sm text-slate-400">暂无题型交叉数据</div>
-          </GlassCard>
+          </BaseCard>
         </div>
       </template>
     </div>
   </div>
+  </ContentPanel>
 </template>
 
 <style scoped>
diff --git a/frontend/src/components/ErrorBank.vue b/frontend/src/views/app/ErrorBankView.vue
similarity index 68%
rename from frontend/src/components/ErrorBank.vue
rename to frontend/src/views/app/ErrorBankView.vue
index 61ddb02d..51064b4d 100644
--- a/frontend/src/components/ErrorBank.vue
+++ b/frontend/src/views/app/ErrorBankView.vue
@@ -1,33 +1,51 @@
 <script setup>
-import { ref, reactive, computed, watch, nextTick, onBeforeUnmount } from 'vue'
-import * as api from '../api.js'
-import { typesetMath as _typesetMath } from '../utils.js'
-import CustomSelect from './CustomSelect.vue'
-import DateRangePicker from './DateRangePicker.vue'
-import GlassCard from './GlassCard.vue'
-import PageHeader from './PageHeader.vue'
-import SearchInput from './SearchInput.vue'
-import QuestionItem from './QuestionItem.vue'
-import QuestionItemSkeleton from './QuestionItemSkeleton.vue'
-import EditNoteDialog from './EditNoteDialog.vue'
-import SelectionPanel from './SelectionPanel.vue'
-import GlassButton from './GlassButton.vue'
-
-const props = defineProps({
-  theme: { type: String, default: 'light' },
-  visible: { type: Boolean, default: false },
-})
-
-const emit = defineEmits(['go-workspace', 'push-toast', 'open-image', 'start-chat'])
+import { ref, reactive, computed, watch, nextTick, onMounted, onBeforeUnmount } from 'vue'
+import * as api from '@/api.js'
+import { typesetMath as _typesetMath } from '@/utils.js'
+import { useSelectableList } from '@/composables/useSelectableList.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import ErrorBankFilterPanel from '@/components/workspace/ErrorBankFilterPanel.vue'
+import BaseSelect from '@/components/base/BaseSelect.vue'
+import BaseCard from '@/components/base/BaseCard.vue'
+import SearchInput from '@/components/base/SearchInput.vue'
+import QuestionItem from '@/components/question/QuestionItem.vue'
+import EmptyState from '@/components/base/EmptyState.vue'
+import QuestionItemSkeleton from '@/components/question/QuestionItemSkeleton.vue'
+import EditNoteDialog from '@/components/question/EditNoteDialog.vue'
+import SelectionPanel from '@/components/workspace/SelectionPanel.vue'
+import { useToast } from '@/composables/useToast.js'
+import { useImageModal } from '@/composables/useImageModal.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+import { useChatSession } from '@/composables/useChatSession.js'
+import { useTheme } from '@/composables/useTheme.js'
+
+const { pushToast } = useToast()
+const { openModal } = useImageModal()
+const { currentView } = useWorkspaceNav()
+const { openChat } = useChatSession()
+const { isDark } = useTheme()
+const theme = computed(() => isDark.value ? 'dark' : 'light')
 
 // ---- 筛选条件 ----
+const openFilter = ref('')
+const filterPanelOpen = ref(false)
+
+function toggleFilterPanel() { filterPanelOpen.value = !filterPanelOpen.value }
+
+// 点击外部关闭筛选下拉
+function closeFilters(e) {
+  if (!e.target.closest('.filter-pill') && !e.target.closest('.dropdown-item')) {
+    openFilter.value = ''
+  }
+}
+onMounted(() => document.addEventListener('click', closeFilters))
+onBeforeUnmount(() => document.removeEventListener('click', closeFilters))
+
 const filters = reactive({
   subject: '',
   knowledge_tag: '',
   question_type: '',
   keyword: '',
-  start_date: '',
-  end_date: '',
   review_status: '',
 })
 const page = ref(1)
@@ -48,16 +66,13 @@ const reviewStatusClass = (status) => {
 // ---- 数据 ----
 const items = ref([])
 const total = ref(0)
+const grandTotal = ref(0)
 const totalPages = ref(0)
 const loading = ref(false)
 const subjects = ref([])
 const questionTypes = ref([])
 const tagNames = ref([])
-const selectedIds = reactive(new Set())
-const selectMode = ref(false)
-
-const enterSelectMode = () => { selectMode.value = true }
-const exitSelectMode = () => { selectMode.value = false; selectedIds.clear() }
+const { selectMode, selectedIds, toggleSelectMode, toggleSelect, clearSelection } = useSelectableList()
 
 
 // ---- 知识点多选标签 ----
@@ -80,7 +95,7 @@ const clearTagSelection = () => {
 }
 
 
-const totalText = computed(() => `共收录 ${total.value} 道题目`)
+const totalText = computed(() => `共收录 ${grandTotal.value} 道题目`)
 
 // ---- 查询 ----
 let debounceTimer = null
@@ -92,16 +107,15 @@ const doQuery = async () => {
     if (filters.knowledge_tag) params.knowledge_tag = filters.knowledge_tag
     if (filters.question_type) params.question_type = filters.question_type
     if (filters.keyword) params.keyword = filters.keyword
-    if (filters.start_date) params.start_date = filters.start_date
-    if (filters.end_date) params.end_date = filters.end_date
     if (filters.review_status) params.review_status = filters.review_status
 
     const data = await api.fetchErrorBank(params)
     items.value = data.items || []
     total.value = data.total || 0
+    grandTotal.value = data.grand_total ?? data.total ?? 0
     totalPages.value = data.total_pages || 0
   } catch (e) {
-    emit('push-toast', 'error', e instanceof Error ? e.message : String(e))
+    pushToast('error', e instanceof Error ? e.message : String(e))
   } finally {
     loading.value = false
     typesetMath()
@@ -113,7 +127,7 @@ const debouncedQuery = () => {
   debounceTimer = setTimeout(() => { page.value = 1; doQuery() }, 300)
 }
 
-watch(() => [filters.subject, filters.knowledge_tag, filters.question_type, filters.start_date, filters.end_date, filters.review_status], debouncedQuery)
+watch(() => [filters.subject, filters.knowledge_tag, filters.question_type, filters.review_status], debouncedQuery)
 watch(() => filters.keyword, () => {
   clearTimeout(debounceTimer)
   debounceTimer = setTimeout(() => { page.value = 1; doQuery() }, 500)
@@ -132,14 +146,12 @@ const goPage = (p) => {
   doQuery()
 }
 
-const toggleSelect = (id) => { selectedIds.has(id) ? selectedIds.delete(id) : selectedIds.add(id) }
-const clearSelection = () => { selectedIds.clear() }
 
 const doExport = async () => {
   if (!selectedIds.size) return
   try {
     const data = await api.exportFromDb(Array.from(selectedIds))
-    emit('push-toast', 'success', '错题本导出成功')
+    pushToast('success', '错题本导出成功')
     let filename = 'wrongbook.md'
     if (data.output_path) {
       const parts = String(data.output_path).split(/[/\\]/)
@@ -150,7 +162,7 @@ const doExport = async () => {
     a.download = filename
     document.body.appendChild(a); a.click(); a.remove()
   } catch (e) {
-    emit('push-toast', 'error', '导出失败: ' + (e instanceof Error ? e.message : String(e)))
+    pushToast('error', '导出失败: ' + (e instanceof Error ? e.message : String(e)))
   }
 }
 
@@ -214,9 +226,9 @@ const onDialogSave = async (draft) => {
       dialogQuestion.value.user_answer = draft
     }
     dialogOpen.value = false
-    emit('push-toast', 'success', '已保存')
+    pushToast('success', '已保存')
   } catch (e) {
-    emit('push-toast', 'error', '保存失败')
+    pushToast('error', '保存失败')
   } finally {
     dialogSaving.value = false
   }
@@ -226,9 +238,9 @@ const quickMarkStatus = async (q, status) => {
   try {
     const data = await api.updateReviewStatus(q.id, status)
     q.review_status = data.review_status
-    emit('push-toast', 'success', `已标记为「${status}」`)
+    pushToast('success', `已标记为「${status}」`)
   } catch (e) {
-    emit('push-toast', 'error', '更新状态失败')
+    pushToast('error', '更新状态失败')
   }
 }
 
@@ -239,9 +251,9 @@ const doDelete = async (q) => {
     items.value = items.value.filter(x => x.id !== q.id)
     total.value = Math.max(0, total.value - 1)
     selectedIds.delete(q.id)
-    emit('push-toast', 'success', '题目已删除')
+    pushToast('success', '题目已删除')
   } catch (e) {
-    emit('push-toast', 'error', '删除失败')
+    pushToast('error', '删除失败')
   }
 }
 
@@ -263,7 +275,8 @@ const pageButtons = computed(() => {
 })
 
 const refreshTags = async () => {
-  tagNames.value = await api.fetchTagNames(filters.subject || undefined)
+  const raw = await api.fetchTagNames(filters.subject || undefined)
+  tagNames.value = [...new Set(raw)]
 }
 
 const scrollContainerRef = ref(null)
@@ -278,7 +291,7 @@ const loadFilters = async () => {
     questionTypes.value = qt
     await refreshTags()
   } catch (e) {
-    emit('push-toast', 'error', '加载筛选项失败')
+    pushToast('error', '加载筛选项失败')
   }
 }
 
@@ -296,9 +309,12 @@ const reloadTags = async () => {
 
 watch(() => filters.subject, () => { reloadTags() })
 
-watch(() => props.visible, (v) => { if (v) { loadFilters(); doQuery() } }, { immediate: true })
+onMounted(() => { loadFilters(); doQuery() })
 
-defineExpose({ refresh: doQuery })
+defineExpose({
+  refresh: doQuery,
+  toggleSelectMode,
+})
 
 onBeforeUnmount(() => {
   if (debounceTimer) clearTimeout(debounceTimer)
@@ -306,95 +322,87 @@ onBeforeUnmount(() => {
 </script>
 
 <template>
-  <div ref="scrollContainerRef" @scroll="handleScroll" class="relative h-full overflow-y-auto custom-scrollbar">
-    <div class="container relative z-10 mx-auto max-w-6xl px-4 py-8 sm:px-8">
-      <!-- 页面标题：强化科技质感 -->
-      <PageHeader
-        title="错题库"
-        :subtitle="`${totalText}，支持按学科、知识点筛选`"
-      >
-        <template #extra>
-          <div class="flex items-center gap-4">
-            <GlassButton :icon="selectMode ? 'fa-xmark' : 'fa-file-export'" class="w-[150px]" @click="selectMode ? exitSelectMode() : enterSelectMode()">
-              {{ selectMode ? '退出选择' : '导出题目' }}
-            </GlassButton>
-            <button @click="emit('go-workspace')" class="btn-primary group h-10 px-8 shadow-md shadow-blue-500/20">
-              <i class="fa-solid fa-plus-circle transition-transform group-hover:rotate-90"></i>
-              录入新题目
-            </button>
-          </div>
-        </template>
-      </PageHeader>
-
-      <!-- 搜索控制台 -->
-      <div class="relative z-20 mb-8 space-y-6">
-        <!-- 第一行：关键词 + 学科 + 题型 + 复习状态 -->
-        <div class="grid grid-cols-1 gap-6 sm:grid-cols-2 lg:grid-cols-4">
-          <SearchInput v-model="filters.keyword" label="内容检索" placeholder="搜索题目关键词..." />
-          <CustomSelect v-model="filters.subject" :options="subjects" label="学科" placeholder="全部学科" />
-          <CustomSelect v-model="filters.question_type" :options="questionTypes" label="题型" placeholder="全部题型" />
-          <CustomSelect v-model="filters.review_status" :options="['待复习', '复习中', '已掌握']" label="复习状态" placeholder="全部状态" />
+  <ContentPanel title="错题库">
+    <template #toolbar>
+      <button @click="currentView = 'workspace'" class="flex h-7 items-center gap-1.5 rounded-md border border-white/[0.06] bg-white/[0.03] px-2.5 text-xs font-medium text-[#8a8f98] hover:bg-white/[0.06] hover:text-[#d0d6e0] transition-colors" title="录入新题目">
+        <i class="fa-solid fa-plus text-[10px]"></i> 录入
+      </button>
+    </template>
+    <template v-if="filterPanelOpen" #sidebar>
+      <ErrorBankFilterPanel
+        :filters="filters"
+        :subjects="subjects"
+        :question-types="questionTypes"
+        :tag-names="tagNames"
+        :selected-tags="selectedTags"
+        @close="filterPanelOpen = false"
+        @toggle-tag="toggleTagSelect"
+      />
+    </template>
+  <div ref="scrollContainerRef" @scroll="handleScroll" class="relative h-full overflow-y-auto custom-scrollbar flex flex-col">
+    <div class="relative z-10 flex-1 flex flex-col">
+      <!-- 搜索 + 操作按钮 + 已激活筛选 pills -->
+      <div class="relative z-20 mb-4 flex items-center gap-2 flex-wrap">
+        <!-- 搜索框 -->
+        <div class="relative">
+          <i class="fa-solid fa-magnifying-glass pointer-events-none absolute left-3 top-1/2 -translate-y-1/2 text-xs text-[#62666d]"></i>
+          <input
+            v-model="filters.keyword"
+            type="text"
+            placeholder="搜索题目..."
+            class="h-8 w-52 rounded-md border border-white/[0.08] bg-white/[0.02] pl-8 pr-3 text-xs font-medium text-[#f7f8f8] placeholder-[#62666d] outline-none transition-colors hover:border-white/[0.12] focus:border-white/[0.15]"
+          />
         </div>
 
-        <!-- 第二行：日期范围 + 重置筛选 -->
-        <div class="flex items-center gap-3">
-          <DateRangePicker
-            :start-date="filters.start_date"
-            :end-date="filters.end_date"
-            label="时间跨度"
-            @update:start-date="filters.start_date = $event"
-            @update:end-date="filters.end_date = $event"
-          />
-          <button @click="resetFilters" title="重置筛选"
-            class="mt-6 inline-flex h-11 w-11 shrink-0 items-center justify-center rounded-xl border border-slate-200/60 bg-white/60 text-slate-500 backdrop-blur-xl transition-all hover:border-slate-300 hover:bg-white/80 hover:text-slate-700 dark:border-white/10 dark:bg-white/[0.03] dark:text-slate-400 dark:hover:border-white/20 dark:hover:bg-white/[0.06] dark:hover:text-slate-200">
-            <i class="fa-solid fa-arrow-rotate-right text-sm"></i>
+        <!-- 操作按钮（推到右侧） -->
+        <div class="ml-auto flex items-center gap-1">
+          <button @click="toggleFilterPanel" class="flex h-7 w-7 items-center justify-center rounded-md border border-white/[0.06] bg-white/[0.03] transition-colors" :class="filterPanelOpen ? 'bg-white/[0.08] text-[#f7f8f8] border-white/[0.12]' : 'text-[#8a8f98] hover:bg-white/[0.06] hover:text-[#d0d6e0]'" title="筛选设置">
+            <i class="fa-solid fa-sliders text-xs"></i>
+          </button>
+          <button @click="toggleSelectMode" class="flex h-7 w-7 items-center justify-center rounded-md border border-white/[0.06] bg-white/[0.03] transition-colors" :class="selectMode ? 'bg-white/[0.08] text-[#f7f8f8] border-white/[0.12]' : 'text-[#8a8f98] hover:bg-white/[0.06] hover:text-[#d0d6e0]'" title="导出题目">
+            <i class="fa-solid fa-file-export text-xs"></i>
           </button>
         </div>
 
-        <!-- 知识点多选标签 -->
-        <Transition
-          enter-active-class="transition-all duration-300 ease-out"
-          leave-active-class="transition-all duration-200 ease-in"
-          enter-from-class="opacity-0 -translate-y-1"
-          leave-to-class="opacity-0 -translate-y-1"
-        >
-          <div v-if="tagNames.length">
-            <label class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500 dark:text-slate-500">知识点标签</label>
-            <div class="no-scrollbar max-h-[128px] overflow-y-auto pr-2 flex flex-wrap gap-2 py-1">
-              <button v-for="(tag, i) in tagNames" :key="tag"
-                @click="toggleTagSelect(tag)"
-                class="rounded-xl px-3 py-1.5 text-xs font-bold transition-all"
-                :style="{ animationDelay: `${i * 20}ms` }"
-                :class="['tag-fade-in', selectedTags.has(tag)
-                  ? 'bg-indigo-500 text-white shadow-md shadow-indigo-500/20 dark:bg-indigo-500 dark:shadow-indigo-500/20'
-                  : 'border border-slate-200/60 bg-white/60 text-slate-600 hover:border-indigo-300 hover:text-indigo-600 dark:border-white/10 dark:bg-white/5 dark:text-slate-400 dark:hover:border-indigo-500/30 dark:hover:text-indigo-300']"
-              >
-                <i v-if="selectedTags.has(tag)" class="fa-solid fa-check mr-1 text-xs"></i>
-                {{ tag }}
-              </button>
-              <button v-if="selectedTags.size" @click="clearTagSelection"
-                class="rounded-xl px-3 py-1.5 text-xs font-bold text-slate-400 hover:text-rose-500 dark:text-slate-500 dark:hover:text-rose-400 transition-colors">
-                <i class="fa-solid fa-xmark mr-1"></i>清除
-              </button>
-            </div>
-          </div>
-        </Transition>
+        <!-- 已激活的筛选 pills（换行显示） -->
+        <button v-if="filters.subject" @click="filters.subject = ''" class="filter-pill filter-pill--active">
+          {{ filters.subject }} <i class="fa-solid fa-xmark text-[8px] ml-1"></i>
+        </button>
+        <button v-if="filters.question_type" @click="filters.question_type = ''" class="filter-pill filter-pill--active">
+          {{ filters.question_type }} <i class="fa-solid fa-xmark text-[8px] ml-1"></i>
+        </button>
+        <button v-if="filters.review_status" @click="filters.review_status = ''" class="filter-pill filter-pill--active">
+          {{ filters.review_status }} <i class="fa-solid fa-xmark text-[8px] ml-1"></i>
+        </button>
+        <template v-for="tag in selectedTags" :key="tag">
+          <button @click="toggleTagSelect(tag)" class="filter-pill filter-pill--active">
+            {{ tag }} <i class="fa-solid fa-xmark text-[8px] ml-1"></i>
+          </button>
+        </template>
+        <button
+          v-if="filters.subject || filters.question_type || filters.review_status || selectedTags.size"
+          @click="filters.subject = ''; filters.question_type = ''; filters.review_status = ''; clearTagSelection()"
+          class="text-xs text-[#62666d] hover:text-rose-400 transition-colors"
+        >清除筛选</button>
       </div>
 
 
       <!-- 列表区 -->
-      <div class="relative">
+      <div class="relative flex-1 flex flex-col">
         <!-- 首次加载：无旧数据时显示骨架 -->
         <QuestionItemSkeleton v-if="loading && !items.length" />
 
         <!-- 空状态 -->
-        <div v-else-if="!loading && !items.length" class="flex flex-col items-center justify-center rounded-[2.5rem] border-2 border-dashed border-slate-200 bg-slate-50/50 py-32 dark:border-white/5 dark:bg-white/5">
-          <div class="mb-8 flex h-24 w-24 items-center justify-center rounded-3xl bg-white shadow-md dark:bg-slate-900">
-            <i class="fa-solid fa-box-open text-4xl text-slate-300 dark:text-slate-700"></i>
-          </div>
-          <p class="text-xl font-black text-slate-900 dark:text-white">暂无匹配记录</p>
-          <p class="mt-2 text-sm font-medium text-slate-500">调整筛选条件，或者开始新的录入</p>
-        </div>
+        <EmptyState
+          v-else-if="!loading && !items.length"
+          icon="fa-solid fa-layer-group"
+          title="暂无匹配记录"
+          description="调整筛选条件，或者开始新的录入"
+        >
+          <button @click="currentView = 'workspace'" class="inline-flex items-center gap-2 rounded-md brand-btn px-4 py-2 text-sm font-medium text-[#f7f8f8]">
+            <i class="fa-solid fa-plus"></i> 录入新题目
+          </button>
+        </EmptyState>
 
         <!-- 列表（有旧数据时保留，遮罩覆盖） -->
         <div v-else class="space-y-4">
@@ -424,7 +432,7 @@ onBeforeUnmount(() => {
               >
                 <div v-if="hoverMenuId === question.id" :style="hoverMenuStyle"
                   @mouseenter="onMenuContentEnter" @mouseleave="onMenuContentLeave"
-                  class="w-44 overflow-hidden rounded-2xl border border-slate-200/60 bg-white/95 p-1.5 shadow-xl backdrop-blur-3xl dark:border-white/10 dark:bg-[#12121A]/90 dark:bg-gradient-to-b dark:from-white/[0.08] dark:to-transparent dark:shadow-[0_20px_50px_rgba(0,0,0,0.6)]">
+                  class="w-44 overflow-hidden rounded-2xl border border-slate-200/60 bg-white/95 p-1.5 shadow-xl dark:border-white/10 dark:bg-[#12121A]/90 dark:bg-gradient-to-b dark:from-white/[0.08] dark:to-transparent dark:shadow-[0_20px_50px_rgba(0,0,0,0.6)]">
                   <div class="px-3 pb-1 pt-2 text-[11px] font-black uppercase tracking-widest text-slate-400 dark:text-slate-500">复习状态</div>
                   <button @click.stop="quickMarkStatus(question, '待复习')"
                     class="flex w-full items-center gap-2 rounded-xl px-3 py-2 text-left text-sm font-bold transition-all"
@@ -454,7 +462,7 @@ onBeforeUnmount(() => {
                     class="flex w-full items-center gap-2 rounded-xl px-3 py-2 text-left text-sm font-bold text-slate-600 transition-all hover:bg-blue-500/10 hover:text-blue-600 dark:text-slate-300 dark:hover:bg-blue-500/20 dark:hover:text-blue-400">
                     <i class="fa-solid fa-note-sticky text-xs"></i>记笔记
                   </button>
-                  <button @click.stop="emit('start-chat', question); hoverMenuId = null"
+                  <button @click.stop="openChat(question); hoverMenuId = null"
                     class="flex w-full items-center gap-2 rounded-xl px-3 py-2 text-left text-sm font-bold text-slate-600 transition-all hover:bg-indigo-500/10 hover:text-indigo-600 dark:text-slate-300 dark:hover:bg-indigo-500/20 dark:hover:text-indigo-400">
                     <i class="fa-solid fa-wand-magic-sparkles text-xs"></i>AI 辅导
                   </button>
@@ -474,7 +482,7 @@ onBeforeUnmount(() => {
         <!-- 遮罩：筛选/翻页时覆盖旧列表，布局不跳动 -->
         <Transition enter-active-class="transition-opacity duration-150" leave-active-class="transition-opacity duration-150" enter-from-class="opacity-0" leave-to-class="opacity-0">
           <div v-if="loading && items.length"
-            class="absolute inset-0 z-10 flex items-center justify-center rounded-2xl bg-white/60 backdrop-blur-sm dark:bg-black/30">
+            class="absolute inset-0 z-10 flex items-center justify-center rounded-2xl bg-white/60 dark:bg-black/30">
             <div class="h-6 w-6 animate-spin rounded-full border-2 border-slate-300 border-t-slate-600 dark:border-white/20 dark:border-t-white/60"></div>
           </div>
         </Transition>
@@ -485,7 +493,7 @@ onBeforeUnmount(() => {
         <button @click="goPage(page - 1)" :disabled="page <= 1" class="inline-flex h-10 w-10 items-center justify-center rounded-xl border border-slate-200/60 bg-white/60 text-slate-700 shadow-sm transition-all hover:bg-white disabled:cursor-not-allowed disabled:opacity-30 dark:border-white/10 dark:bg-white/5 dark:text-slate-300">
           <i class="fa-solid fa-chevron-left text-sm"></i>
         </button>
-        <div class="flex items-center gap-2 rounded-2xl bg-white/50 p-1.5 shadow-sm backdrop-blur-md dark:bg-white/5">
+        <div class="flex items-center gap-2 rounded-2xl bg-white/50 p-1.5 shadow-sm dark:bg-white/5">
           <template v-for="(p, i) in pageButtons" :key="i">
             <span v-if="p === '...'" class="flex w-8 justify-center font-bold text-slate-400">...</span>
             <button v-else @click="goPage(p)" 
@@ -522,6 +530,7 @@ onBeforeUnmount(() => {
       @save="onDialogSave"
     />
   </div>
+  </ContentPanel>
 </template>
 
 <style scoped>
diff --git a/frontend/src/views/app/NoteView.vue b/frontend/src/views/app/NoteView.vue
new file mode 100644
index 00000000..84c42fe0
--- /dev/null
+++ b/frontend/src/views/app/NoteView.vue
@@ -0,0 +1,340 @@
+<script setup>
+import { ref, watch, computed, nextTick, onMounted } from 'vue'
+import * as api from '@/api.js'
+import { renderMarkdown, typesetMath } from '@/utils.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import BaseCard from '@/components/base/BaseCard.vue'
+import BaseGhostButton from '@/components/base/BaseGhostButton.vue'
+import SearchInput from '@/components/base/SearchInput.vue'
+import BaseSelect from '@/components/base/BaseSelect.vue'
+import EmptyState from '@/components/base/EmptyState.vue'
+import { useToast } from '@/composables/useToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import { useTheme } from '@/composables/useTheme.js'
+
+const { pushToast } = useToast()
+const { selectedProvider, selectedModel } = useSystemStatus()
+const { isDark } = useTheme()
+const theme = computed(() => isDark.value ? 'dark' : 'light')
+
+const noteContentRef = ref(null)
+
+// ---- 状态 ----
+const notes = ref([])
+const total = ref(0)
+const page = ref(1)
+const pageSize = ref(10)
+const totalPages = computed(() => Math.ceil(total.value / pageSize.value))
+const loading = ref(false)
+const creating = ref(false)
+const createProgress = ref(0)
+
+// 筛选
+const filterSubject = ref('')
+const filterTag = ref('')
+const filterKeyword = ref('')
+const subjects = ref([])
+const tagNames = ref([])
+
+// 加载筛选选项
+async function loadFilterOptions() {
+  try {
+    const subjectData = await api.fetchSubjects()
+    subjects.value = subjectData.subjects || []
+  } catch (_) {}
+}
+
+// 详情/编辑
+const selectedNote = ref(null)
+const editing = ref(false)
+const editTitle = ref('')
+const editContent = ref('')
+
+// ---- 加载笔记列表 ----
+async function loadNotes() {
+  loading.value = true
+  try {
+    const data = await api.fetchNotes({
+      page: page.value,
+      limit: pageSize.value,
+      subject: filterSubject.value || undefined,
+      knowledge_tag: filterTag.value || undefined,
+      keyword: filterKeyword.value || undefined,
+    })
+    notes.value = data.items
+    total.value = data.total
+  } catch (e) {
+    pushToast('error', e.message)
+  } finally {
+    loading.value = false
+  }
+}
+
+onMounted(() => { loadNotes(); loadFilterOptions() })
+watch([page, filterSubject, filterTag], () => { page.value === 1 ? loadNotes() : (page.value = 1) })
+
+let keywordTimer = null
+watch(filterKeyword, () => {
+  clearTimeout(keywordTimer)
+  keywordTimer = setTimeout(() => { page.value = 1; loadNotes() }, 500)
+})
+
+// ---- 上传笔记 ----
+const fileInput = ref(null)
+
+function triggerUpload() {
+  fileInput.value?.click()
+}
+
+function handleFiles(e) {
+  const files = e.target.files
+  if (!files?.length) return
+
+  creating.value = true
+  createProgress.value = 0
+
+  const formData = new FormData()
+  for (const f of files) formData.append('files', f)
+  formData.append('model_provider', selectedProvider.value)
+  if (selectedModel.value) formData.append('model_name', selectedModel.value)
+
+  api.createNote(formData, {
+    onProgress: (ratio) => { createProgress.value = Math.round(ratio * 50) },
+    onSuccess: (data) => {
+      creating.value = false
+      createProgress.value = 100
+      pushToast('success', '笔记整理完成')
+      loadNotes()
+      // 自动打开新创建的笔记
+      if (data.note) selectedNote.value = data.note
+    },
+    onError: (msg) => {
+      creating.value = false
+      emit('push-toast', 'error', msg)
+    },
+  })
+
+  // 模拟 OCR + LLM 处理进度
+  let fakeProgress = 50
+  const timer = setInterval(() => {
+    if (!creating.value) { clearInterval(timer); return }
+    fakeProgress = Math.min(fakeProgress + 2, 95)
+    createProgress.value = fakeProgress
+  }, 500)
+
+  e.target.value = ''
+}
+
+// ---- 详情 ----
+async function openNote(note) {
+  try {
+    const full = await api.fetchNote(note.id)
+    selectedNote.value = full
+    // 等 DOM 完全更新后渲染 LaTeX 公式（nextTick + 延迟，确保过渡动画完成）
+    await nextTick()
+    setTimeout(() => {
+      if (noteContentRef.value) typesetMath(noteContentRef.value)
+    }, 100)
+  } catch (e) {
+    pushToast('error', e.message)
+  }
+}
+
+function closeDetail() {
+  selectedNote.value = null
+  editing.value = false
+}
+
+// ---- 编辑 ----
+function startEdit() {
+  editTitle.value = selectedNote.value.title
+  editContent.value = selectedNote.value.content_markdown
+  editing.value = true
+}
+
+async function saveEdit() {
+  try {
+    const updated = await api.updateNote(selectedNote.value.id, {
+      title: editTitle.value,
+      content_markdown: editContent.value,
+    })
+    selectedNote.value = updated
+    editing.value = false
+    pushToast('success', '已保存')
+    loadNotes()
+    await nextTick()
+    setTimeout(() => {
+      if (noteContentRef.value) typesetMath(noteContentRef.value)
+    }, 100)
+  } catch (e) {
+    pushToast('error', e.message)
+  }
+}
+
+// ---- 删除 ----
+async function doDelete(noteId) {
+  if (!confirm('确认删除这条笔记？')) return
+  try {
+    await api.deleteNote(noteId)
+    pushToast('success', '已删除')
+    if (selectedNote.value?.id === noteId) closeDetail()
+    loadNotes()
+  } catch (e) {
+    pushToast('error', e.message)
+  }
+}
+</script>
+
+<template>
+  <ContentPanel title="笔记库">
+    <template #toolbar>
+      <button @click="triggerUpload" class="flex h-7 items-center gap-1.5 rounded-md border border-white/[0.06] bg-white/[0.03] px-2.5 text-xs font-medium text-[#8a8f98] hover:bg-white/[0.06] hover:text-[#d0d6e0] transition-colors" title="上传笔记">
+        <i class="fa-solid fa-upload text-[10px]"></i> 上传笔记
+      </button>
+    </template>
+  <div class="relative h-full overflow-y-auto custom-scrollbar">
+    <div class="container relative z-10 mx-auto max-w-6xl px-4 py-8 sm:px-8">
+      
+      <!-- List View -->
+      <div v-if="!selectedNote">
+        <input ref="fileInput" type="file" multiple accept="image/*" class="hidden" @change="handleFiles" />
+
+        <!-- 筛选栏（对齐错题库风格） -->
+        <div class="relative z-20 mb-8">
+          <div class="grid grid-cols-1 gap-4 sm:grid-cols-4">
+            <SearchInput v-model="filterKeyword" label="内容检索" placeholder="搜索笔记关键词..." />
+            <BaseSelect v-model="filterSubject" :options="subjects" label="学科" placeholder="全部学科" />
+            <BaseSelect v-model="filterTag" :options="tagNames" label="知识点" placeholder="全部知识点" />
+            <div>
+              <label class="mb-1.5 block text-xs font-medium text-[#62666d]">统计</label>
+              <div class="flex h-9 items-center rounded-md border border-white/[0.08] bg-white/[0.02] px-3 text-sm text-[#8a8f98]">
+                共 {{ total }} 条笔记
+              </div>
+            </div>
+          </div>
+
+          <!-- 进度条 -->
+          <div v-if="creating" class="mt-4">
+            <div class="h-1 rounded-full bg-white/[0.06]">
+              <div class="h-full rounded-full bg-[rgb(129,115,223)] transition-all duration-300" :style="{ width: createProgress + '%' }"></div>
+            </div>
+            <p class="mt-1 text-xs text-[#62666d]">OCR 识别 + AI 整理中... {{ createProgress }}%</p>
+          </div>
+        </div>
+
+        <!-- Notes Grid -->
+        <div class="relative">
+          <EmptyState
+            v-if="!loading && notes.length === 0"
+            icon="fa-solid fa-book-open"
+            title="还没有笔记"
+            description="上传手写笔记或板书照片，AI 自动整理为结构化知识点"
+          >
+            <button @click="triggerUpload" class="inline-flex items-center gap-2 rounded-md brand-btn px-4 py-2 text-sm font-medium text-[#f7f8f8]">
+              <i class="fa-solid fa-plus"></i> 上传笔记
+            </button>
+          </EmptyState>
+
+          <div v-else class="grid gap-4 sm:grid-cols-2 lg:grid-cols-3">
+            <div
+              v-for="note in notes"
+              :key="note.id"
+              @click="openNote(note)"
+              class="group cursor-pointer rounded-lg border border-white/[0.06] bg-white/[0.02] p-4 transition-all hover:bg-white/[0.04] hover:border-white/[0.1]"
+            >
+              <h3 class="mb-2 text-sm font-medium text-[#f7f8f8] line-clamp-2">{{ note.title }}</h3>
+              <p class="mb-3 text-xs text-[#8a8f98] line-clamp-3">
+                {{ note.content_markdown?.replace(/[#*`>\-]/g, '').slice(0, 120) }}...
+              </p>
+              <div class="flex flex-wrap items-center gap-2">
+                <span v-if="note.subject" class="rounded-md bg-[rgb(129,115,223)]/10 px-2 py-0.5 text-xs font-medium text-[rgb(145,132,235)]">{{ note.subject }}</span>
+                <span v-for="tag in (note.knowledge_tags || []).slice(0, 3)" :key="tag" class="rounded-md border border-white/[0.06] px-2 py-0.5 text-xs text-[#8a8f98]">{{ tag }}</span>
+                <span class="ml-auto text-xs text-[#62666d]">{{ note.updated_at?.slice(0, 10) }}</span>
+              </div>
+            </div>
+          </div>
+
+          <!-- 分页 -->
+          <div v-if="totalPages > 1" class="mt-8 flex justify-center gap-2">
+            <button
+              v-for="p in totalPages"
+              :key="p"
+              @click="page = p"
+              class="size-8 rounded-md text-xs font-medium transition-all"
+              :class="p === page ? 'brand-btn text-[#f7f8f8]' : 'border border-white/[0.06] text-[#8a8f98] hover:bg-white/[0.04]'"
+            >
+              {{ p }}
+            </button>
+          </div>
+        </div>
+      </div>
+
+      <!-- Detail View -->
+      <div v-else>
+        <!-- 详情工具栏 -->
+        <div class="mb-6 flex items-center justify-between">
+          <h3 class="text-base font-medium text-[#f7f8f8]">{{ selectedNote.title }}</h3>
+          <div class="flex items-center gap-2">
+            <button @click="closeDetail" class="inline-flex items-center gap-2 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#d0d6e0] hover:bg-white/[0.05] transition-colors">
+              <i class="fa-solid fa-arrow-left text-[10px]"></i> 返回
+            </button>
+            <button v-if="!editing" @click="startEdit" class="inline-flex items-center gap-2 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#d0d6e0] hover:bg-white/[0.05] transition-colors">
+              <i class="fa-solid fa-pen text-[10px]"></i> 编辑
+            </button>
+            <button @click="doDelete(selectedNote.id)" class="inline-flex items-center gap-2 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-rose-400 hover:bg-rose-500/10 transition-colors">
+              <i class="fa-solid fa-trash text-[10px]"></i> 删除
+            </button>
+          </div>
+        </div>
+
+        <!-- Tags Row -->
+        <div class="mb-6 flex flex-wrap items-center gap-2">
+          <span v-if="selectedNote.subject" class="rounded-full bg-emerald-100 px-3 py-1 text-xs font-bold text-emerald-700 dark:bg-emerald-500/20 dark:text-emerald-300">{{ selectedNote.subject }}</span>
+          <span v-for="tag in selectedNote.knowledge_tags" :key="tag" class="rounded-full bg-slate-100 px-3 py-1 text-xs text-slate-600 dark:bg-white/[0.06] dark:text-slate-300">{{ tag }}</span>
+        </div>
+
+        <BaseCard padding="p-8">
+          <!-- Edit Mode -->
+          <div v-if="editing" class="space-y-4">
+            <div>
+              <label class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500">标题</label>
+              <input v-model="editTitle" class="w-full rounded-xl border border-slate-200/60 bg-white/60 px-4 h-12 text-lg font-bold outline-none focus:border-emerald-300 focus:ring-2 focus:ring-emerald-500/20 dark:border-white/10 dark:bg-white/[0.03] dark:text-white" />
+            </div>
+            <div>
+              <label class="mb-2 block text-[11px] font-black uppercase tracking-widest text-slate-500">内容 (Markdown)</label>
+              <textarea v-model="editContent" rows="15" class="w-full rounded-xl border border-slate-200/60 bg-white/60 p-4 font-mono text-sm leading-relaxed outline-none focus:border-emerald-300 focus:ring-2 focus:ring-emerald-500/20 custom-scrollbar dark:border-white/10 dark:bg-white/[0.03] dark:text-slate-200"></textarea>
+            </div>
+            <div class="flex gap-3 pt-4">
+              <button @click="saveEdit" class="btn-primary h-10 px-8 shadow-md shadow-emerald-500/20">
+                <i class="fa-solid fa-check mr-2"></i>保存修改
+              </button>
+              <BaseGhostButton @click="editing = false">取消</BaseGhostButton>
+            </div>
+          </div>
+
+          <!-- View Mode -->
+          <div v-else ref="noteContentRef">
+            <article class="prose prose-slate max-w-none dark:prose-invert prose-headings:text-slate-900 dark:prose-headings:text-white prose-p:leading-relaxed prose-a:text-emerald-600 prose-pre:bg-slate-50 dark:prose-pre:bg-slate-900 prose-pre:border prose-pre:border-slate-200/60 dark:prose-pre:border-white/10" v-html="renderMarkdown(selectedNote.content_markdown || '')"></article>
+          </div>
+        </BaseCard>
+
+        <!-- Original Images -->
+        <div v-if="selectedNote.source_images?.length && !editing" class="mt-8">
+          <h3 class="mb-4 text-sm font-black uppercase tracking-widest text-slate-500">
+            <i class="fa-solid fa-image mr-2"></i>原始笔记图片
+          </h3>
+          <div class="grid gap-4 sm:grid-cols-2">
+            <BaseCard v-for="(src, idx) in selectedNote.source_images" :key="idx" padding="p-2" class="overflow-hidden">
+              <img
+                :src="'/images/' + src.split('/imgs/').pop() || src"
+                class="w-full rounded-xl cursor-pointer hover:opacity-90 transition-opacity"
+                @error="$event.target.style.display='none'"
+              />
+            </BaseCard>
+          </div>
+        </div>
+      </div>
+    </div>
+  </div>
+  </ContentPanel>
+</template>
diff --git a/frontend/src/components/ReviewView.vue b/frontend/src/views/app/ReviewView.vue
similarity index 82%
rename from frontend/src/components/ReviewView.vue
rename to frontend/src/views/app/ReviewView.vue
index aabc7176..7bf3b6c9 100644
--- a/frontend/src/components/ReviewView.vue
+++ b/frontend/src/views/app/ReviewView.vue
@@ -1,25 +1,31 @@
 <script setup>
-import { ref, reactive, watch, nextTick, onBeforeUnmount, computed } from 'vue'
-import * as api from '../api.js'
-import { typesetMath as _typesetMath } from '../utils.js'
-import QuestionDetailModal from './QuestionDetailModal.vue'
-import AiAnalysisModal from './AiAnalysisModal.vue'
-import CustomSelect from './CustomSelect.vue'
-import PageHeader from './PageHeader.vue'
-import QuestionItem from './QuestionItem.vue'
+import { ref, watch, nextTick, onMounted, onBeforeUnmount, computed } from 'vue'
+import * as api from '@/api.js'
+import { typesetMath as _typesetMath } from '@/utils.js'
+import { useSelectableList } from '@/composables/useSelectableList.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import QuestionDetailModal from '@/components/question/QuestionDetailModal.vue'
+import AiAnalysisModal from '@/components/review/AiAnalysisModal.vue'
+import BaseSelect from '@/components/base/BaseSelect.vue'
+import QuestionItem from '@/components/question/QuestionItem.vue'
+import { useToast } from '@/composables/useToast.js'
+import { useImageModal } from '@/composables/useImageModal.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+import { useChatSession } from '@/composables/useChatSession.js'
+import { useTheme } from '@/composables/useTheme.js'
+
+const { pushToast } = useToast()
+const { openModal } = useImageModal()
+const { currentView } = useWorkspaceNav()
+const { openChat } = useChatSession()
+const { isDark } = useTheme()
+const theme = computed(() => isDark.value ? 'dark' : 'light')
 
 const answerEditId = ref(null)
 const answerEditField = ref('')
 const answerEditDraft = ref('')
 const answerEditSaving = ref(false)
 
-const props = defineProps({
-  theme: { type: String, default: 'light' },
-  visible: { type: Boolean, default: false },
-})
-
-const emit = defineEmits(['go-workspace', 'push-toast', 'open-image', 'start-chat'])
-
 // ---- 学科筛选 ----
 const selectedSubject = ref('')
 const subjects = ref([])
@@ -34,8 +40,7 @@ const detailOpen = ref(false)
 const detailQuestion = ref(null)
 
 // ---- AI 分析 ----
-const selectMode = ref(false)
-const selectedIds = reactive(new Set())
+const { selectMode, selectedIds, toggleSelectMode, toggleSelect, selectAllItems } = useSelectableList()
 const aiModalOpen = ref(false)
 const aiAnalysisResult = ref(null)
 const aiAnalyzing = ref(false)
@@ -57,7 +62,7 @@ const loadReviewItems = async () => {
     reviewItems.value = data.items || []
     reviewTotal.value = data.total || 0
   } catch (e) {
-    emit('push-toast', 'error', '加载待复习题目失败')
+    pushToast('error', '加载待复习题目失败')
   } finally {
     reviewLoading.value = false
     typesetMath()
@@ -81,13 +86,13 @@ const quickMarkStatus = async (q, status) => {
   try {
     const data = await api.updateReviewStatus(q.id, status)
     q.review_status = data.review_status
-    emit('push-toast', 'success', `已标记为「${status}」`)
+    pushToast('success', `已标记为「${status}」`)
     if (status !== '待复习') {
       reviewItems.value = reviewItems.value.filter(x => x.id !== q.id)
       reviewTotal.value = Math.max(0, reviewTotal.value - 1)
     }
   } catch (e) {
-    emit('push-toast', 'error', '更新状态失败')
+    pushToast('error', '更新状态失败')
   }
 }
 
@@ -122,9 +127,9 @@ const saveInlineEdit = async () => {
       q.user_answer = answerEditDraft.value
     }
     answerEditId.value = null
-    emit('push-toast', 'success', '已保存')
+    pushToast('success', '已保存')
   } catch (e) {
-    emit('push-toast', 'error', '保存失败')
+    pushToast('error', '保存失败')
   } finally {
     answerEditSaving.value = false
   }
@@ -139,24 +144,11 @@ const onReviewStatusChanged = (id, status, updatedAt) => {
   }
 }
 
-// ---- AI 分析 ----
-const toggleSelectMode = () => {
-  selectMode.value = !selectMode.value
-  if (!selectMode.value) selectedIds.clear()
-}
-
-const toggleSelect = (id) => {
-  if (selectedIds.has(id)) selectedIds.delete(id)
-  else selectedIds.add(id)
-}
-
-const selectAllReview = () => {
-  for (const q of reviewItems.value) selectedIds.add(q.id)
-}
+const selectAllReview = () => selectAllItems(reviewItems.value)
 
 const startAiAnalysis = async () => {
   if (!selectedIds.size) {
-    emit('push-toast', 'error', '请先选择要分析的题目')
+    pushToast('error', '请先选择要分析的题目')
     return
   }
   aiAnalyzing.value = true
@@ -166,7 +158,7 @@ const startAiAnalysis = async () => {
     const data = await api.requestAiAnalysis(Array.from(selectedIds))
     aiAnalysisResult.value = data.analysis
   } catch (e) {
-    emit('push-toast', 'error', 'AI 分析失败: ' + (e instanceof Error ? e.message : String(e)))
+    pushToast('error', 'AI 分析失败: ' + (e instanceof Error ? e.message : String(e)))
     aiModalOpen.value = false
   } finally {
     aiAnalyzing.value = false
@@ -179,28 +171,19 @@ const closeAiModal = () => {
 }
 
 // ---- 生命周期 ----
-watch(() => props.visible, (v) => {
-  if (v) loadAll()
-}, { immediate: true })
+onMounted(() => { loadAll() })
 </script>
 
 <template>
+  <ContentPanel title="待复习">
+    <template #toolbar>
+      <BaseSelect v-if="subjects.length" v-model="selectedSubject" :options="subjects" placeholder="全部学科" width-class="min-w-[140px]" />
+      <button @click="currentView = 'workspace'" class="btn-primary group h-10 px-8 shadow-md shadow-blue-500/20">
+        <i class="fa-solid fa-plus-circle transition-transform group-hover:rotate-90"></i> 录入新题目
+      </button>
+    </template>
   <div class="relative h-full overflow-y-auto custom-scrollbar">
     <div class="container relative z-10 mx-auto max-w-6xl px-4 py-8 sm:px-8">
-      <!-- 页面标题 -->
-      <PageHeader
-        title="待复习"
-        subtitle="需要巩固的题目，标记掌握后自动归档"
-      >
-        <template #extra>
-          <div class="flex items-center gap-4">
-            <CustomSelect v-if="subjects.length" v-model="selectedSubject" :options="subjects" placeholder="全部学科" width-class="min-w-[140px]" />
-            <button @click="emit('go-workspace')" class="btn-primary group h-10 px-8 shadow-md shadow-blue-500/20">
-              <i class="fa-solid fa-plus-circle transition-transform group-hover:rotate-90"></i> 录入新题目
-            </button>
-          </div>
-        </template>
-      </PageHeader>
 
       <!-- 操作栏 -->
       <div class="mb-6 flex flex-wrap items-center justify-between gap-4">
@@ -283,7 +266,7 @@ watch(() => props.visible, (v) => {
               <button @click="quickMarkStatus(q, '复习中')" class="rounded-xl border border-indigo-200 bg-indigo-50 px-4 py-2 text-xs font-bold text-indigo-600 transition-all hover:bg-indigo-100 dark:border-indigo-500/20 dark:bg-indigo-500/10 dark:text-indigo-400">
                 <i class="fa-solid fa-spinner mr-1"></i>复习中
               </button>
-              <span class="pointer-events-none absolute -top-7 left-1/2 -translate-x-1/2 whitespace-nowrap rounded-md bg-slate-600/90 px-2 py-0.5 text-xs font-semibold text-white opacity-0 shadow-sm backdrop-blur-sm transition-opacity group-hover/tip:opacity-100 dark:bg-slate-700/90">
+              <span class="pointer-events-none absolute -top-7 left-1/2 -translate-x-1/2 whitespace-nowrap rounded-md bg-slate-600/90 px-2 py-0.5 text-xs font-semibold text-white opacity-0 shadow-sm transition-opacity group-hover/tip:opacity-100 dark:bg-slate-700/90">
                 标记为复习中
                 <span class="absolute -bottom-1 left-1/2 -translate-x-1/2 border-[3px] border-transparent border-t-slate-600/90 dark:border-t-slate-700/90"></span>
               </span>
@@ -292,7 +275,7 @@ watch(() => props.visible, (v) => {
               <button @click="quickMarkStatus(q, '已掌握')" class="rounded-xl border border-emerald-500/30 bg-emerald-50/80 px-4 py-2 text-xs font-bold text-emerald-700 transition-all hover:bg-emerald-100 dark:border-emerald-500/20 dark:bg-emerald-500/10 dark:text-emerald-400">
                 <i class="fa-solid fa-circle-check mr-1"></i>已掌握
               </button>
-              <span class="pointer-events-none absolute -top-7 left-1/2 -translate-x-1/2 whitespace-nowrap rounded-md bg-slate-600/90 px-2 py-0.5 text-xs font-semibold text-white opacity-0 shadow-sm backdrop-blur-sm transition-opacity group-hover/tip:opacity-100 dark:bg-slate-700/90">
+              <span class="pointer-events-none absolute -top-7 left-1/2 -translate-x-1/2 whitespace-nowrap rounded-md bg-slate-600/90 px-2 py-0.5 text-xs font-semibold text-white opacity-0 shadow-sm transition-opacity group-hover/tip:opacity-100 dark:bg-slate-700/90">
                 标记为已掌握
                 <span class="absolute -bottom-1 left-1/2 -translate-x-1/2 border-[3px] border-transparent border-t-slate-600/90 dark:border-t-slate-700/90"></span>
               </span>
@@ -307,12 +290,12 @@ watch(() => props.visible, (v) => {
       :open="detailOpen"
       :question="detailQuestion"
       @close="closeDetail"
-      @open-image="(src) => emit('open-image', src)"
+      @open-image="(src) => openModal(src)"
       @deleted="onDeleted"
       @answer-saved="onAnswerSaved"
       @review-status-changed="onReviewStatusChanged"
-      @push-toast="(type, msg) => emit('push-toast', type, msg)"
-      @start-chat="(q) => emit('start-chat', q)"
+      @push-toast="(type, msg) => pushToast(type, msg)"
+      @start-chat="(q) => openChat(q)"
     />
 
     <!-- AI 分析弹窗 -->
@@ -323,5 +306,6 @@ watch(() => props.visible, (v) => {
       @close="closeAiModal"
     />
   </div>
+  </ContentPanel>
 </template>
 
diff --git a/frontend/src/components/SettingsView.vue b/frontend/src/views/app/SettingsView.vue
similarity index 92%
rename from frontend/src/components/SettingsView.vue
rename to frontend/src/views/app/SettingsView.vue
index 0de3870d..a0bc89d7 100644
--- a/frontend/src/components/SettingsView.vue
+++ b/frontend/src/views/app/SettingsView.vue
@@ -1,17 +1,15 @@
 <script setup>
-import { ref, inject, onMounted, watch } from 'vue'
-import { fetchAppConfig, updateAppConfig } from '../api.js'
-import { genId } from '../utils.js'
-import ProviderDialog from './ProviderDialog.vue'
-import ProviderSection from './ProviderSection.vue'
-
-const props = defineProps({
-  visible: { type: Boolean, default: false },
-})
-
-const emit = defineEmits(['saved'])
-
-const pushToast = inject('pushToast', (type, msg) => { console.warn(`[${type}] ${msg}`) })
+import { ref, onMounted, watch } from 'vue'
+import { fetchAppConfig, updateAppConfig } from '@/api.js'
+import { genId } from '@/utils.js'
+import { useToast } from '@/composables/useToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import ProviderDialog from '@/components/settings/ProviderDialog.vue'
+import ProviderSection from '@/components/settings/ProviderSection.vue'
+
+const { pushToast } = useToast()
+const { doFetchStatus } = useSystemStatus()
 
 const loading = ref(true)
 const saving = ref(false)
@@ -140,7 +138,7 @@ const saveConfig = async () => {
     payload.active_paddleocr_id = activePaddleocrId.value
 
     await updateAppConfig(payload)
-    emit('saved')
+    doFetchStatus()
   } catch (e) {
     pushToast('error', '保存失败: ' + (e instanceof Error ? e.message : String(e)))
   } finally {
@@ -235,19 +233,16 @@ const onDialogConfirm = async (formData) => {
   } catch { /* saveConfig 内部已 toast error */ }
 }
 
-onMounted(() => { if (props.visible) loadConfig() })
-watch(() => props.visible, (v) => { if (v) loadConfig() })
+onMounted(() => { loadConfig() })
 </script>
 
 <template>
+  <ContentPanel title="系统设置">
   <div class="relative h-full overflow-y-auto">
     <div class="container relative z-10 mx-auto max-w-3xl px-4 py-8 sm:px-8">
-      <!-- 页面标题 -->
+      <!-- 页面说明 -->
       <div class="mb-8 pl-2 sm:pl-0">
-        <h2 class="text-2xl font-bold tracking-tight text-slate-900 sm:text-3xl dark:text-white">
-          系统设置
-        </h2>
-        <p class="mt-2 text-sm font-medium text-slate-500 dark:text-slate-400">配置 AI 模型供应商和 OCR 服务连接参数，修改即时生效。</p>
+        <p class="text-sm font-medium text-slate-500 dark:text-slate-400">配置 AI 模型供应商和 OCR 服务连接参数，修改即时生效。</p>
       </div>
 
       <!-- 加载中 -->
@@ -309,4 +304,5 @@ watch(() => props.visible, (v) => { if (v) loadConfig() })
       @confirm="onDialogConfirm"
     />
   </div>
+  </ContentPanel>
 </template>
diff --git a/frontend/src/components/SplitHistory.vue b/frontend/src/views/app/SplitHistoryView.vue
similarity index 60%
rename from frontend/src/components/SplitHistory.vue
rename to frontend/src/views/app/SplitHistoryView.vue
index 518b9dd8..2b4d6552 100644
--- a/frontend/src/components/SplitHistory.vue
+++ b/frontend/src/views/app/SplitHistoryView.vue
@@ -1,7 +1,7 @@
 <script setup>
 import { ref, watch, nextTick } from 'vue'
-import * as api from '../api.js'
-import { isHtml, sanitizeHtml, typesetMath as _typesetMath } from '../utils.js'
+import * as api from '@/api.js'
+import { isHtml, sanitizeHtml, typesetMath as _typesetMath } from '@/utils.js'
 
 const props = defineProps({
   theme: { type: String, default: 'light' },
@@ -42,7 +42,10 @@ const openDetail = async (record) => {
   selectedIds.value = new Set()
   try {
     const detail = await api.fetchSplitRecordDetail(record.id)
-    activeQuestions.value = detail.questions || []
+    // 兼容旧历史记录（无 uid）：按索引补齐
+    const qs = detail.questions || []
+    qs.forEach((q, i) => { if (q.uid == null) q.uid = `hist_${i}` })
+    activeQuestions.value = qs
   } catch (e) {
     emit('push-toast', 'error', '加载记录详情失败')
     activeRecord.value = null
@@ -68,7 +71,7 @@ const toggleSelect = (qid) => {
 }
 
 const selectAllQuestions = () => {
-  selectedIds.value = new Set(activeQuestions.value.map(q => q.question_id))
+  selectedIds.value = new Set(activeQuestions.value.map(q => q.uid))
 }
 
 const deselectAllQuestions = () => {
@@ -78,7 +81,7 @@ const deselectAllQuestions = () => {
 const loadToWorkspace = () => {
   if (!activeQuestions.value.length) return
   const qs = selectedIds.value.size > 0
-    ? activeQuestions.value.filter(q => selectedIds.value.has(q.question_id))
+    ? activeQuestions.value.filter(q => selectedIds.value.has(q.uid))
     : activeQuestions.value
   emit('load-record', qs, activeRecord.value)
 }
@@ -146,130 +149,52 @@ defineExpose({ refresh: loadRecords })
 
 <template>
   <div v-show="visible" class="flex h-full flex-col overflow-hidden">
-    <!-- 背景光晕 -->
-    <div class="pointer-events-none absolute inset-0 z-0 overflow-hidden">
-      <div class="animate-blob absolute -top-[15%] right-[-5%] h-[45vw] w-[45vw] rounded-full bg-violet-300/[0.10] mix-blend-multiply blur-[120px] dark:bg-violet-600/15 dark:mix-blend-screen"></div>
-      <div class="animate-blob animation-delay-4000 absolute -bottom-[10%] left-[-8%] h-[40vw] w-[40vw] rounded-full bg-blue-200/[0.12] mix-blend-multiply blur-[100px] dark:bg-blue-500/10 dark:mix-blend-screen"></div>
-    </div>
-
-    <div class="container relative z-10 mx-auto flex h-full max-w-6xl flex-col px-4 py-4 sm:px-8 sm:py-6">
+    <div class="relative flex h-full flex-col">
       <!-- 页头 -->
-      <div class="mb-6 flex flex-col gap-2 sm:flex-row sm:items-center sm:justify-between shrink-0">
-        <div class="flex items-center gap-4">
-          <button
-            @click="emit('go-workspace')"
-            class="group flex h-10 w-10 items-center justify-center rounded-xl border border-slate-200/60 bg-white/60 text-slate-500 backdrop-blur-md transition-all hover:border-blue-500/50 hover:bg-white hover:text-blue-600 dark:border-white/10 dark:bg-white/5 dark:text-slate-400 dark:hover:border-indigo-500/50 dark:hover:text-indigo-300"
-          >
-            <i class="fa-solid fa-arrow-left-long transition-transform group-hover:-translate-x-1"></i>
-          </button>
-          <div>
-            <div class="mb-1 flex items-center gap-2">
-              <span class="inline-flex items-center rounded-full bg-violet-100 px-2 py-0.5 text-[9px] font-bold uppercase tracking-wider text-violet-700 dark:bg-violet-500/20 dark:text-violet-300">
-                History
-              </span>
-              <span class="h-1 w-1 rounded-full bg-slate-300 dark:bg-slate-600"></span>
-              <span class="text-[9px] font-bold uppercase tracking-wider text-slate-400 dark:text-slate-500">最近 20 条</span>
-            </div>
-            <h2 class="text-2xl font-extrabold tracking-tight text-slate-900 sm:text-3xl dark:text-white">分割历史</h2>
-          </div>
-        </div>
+      <div class="flex items-center justify-between px-4 py-3 border-b border-white/[0.05] shrink-0">
+        <span class="text-xs font-medium text-[#f7f8f8]">分割历史</span>
         <button
           @click="loadRecords"
           :disabled="loading"
-          class="inline-flex items-center gap-2 rounded-xl border border-slate-200/60 bg-white/60 px-4 py-2.5 text-sm font-bold text-slate-600 backdrop-blur-md transition-all hover:border-violet-500/40 hover:bg-white hover:text-violet-600 disabled:opacity-50 dark:border-white/10 dark:bg-white/5 dark:text-slate-300 dark:hover:border-violet-500/40 dark:hover:text-violet-300"
+          class="text-xs text-[#8a8f98] hover:text-[#d0d6e0] transition-colors disabled:opacity-50"
         >
-          <i class="fa-solid fa-arrows-rotate text-sm" :class="{ 'animate-spin': loading }"></i>
-          刷新
+          <i class="fa-solid fa-arrows-rotate text-[10px]" :class="{ 'animate-spin': loading }"></i>
         </button>
       </div>
 
       <!-- 主体 -->
-      <div class="flex-1 overflow-y-auto pr-1 custom-scrollbar">
+      <div class="flex-1 overflow-y-auto custom-scrollbar">
 
         <!-- 加载状态 -->
-        <div v-if="loading && !records.length" class="flex flex-col items-center justify-center py-20">
-          <div class="mb-4 h-10 w-10 animate-spin rounded-full border-[3px] border-slate-200 border-t-violet-500 dark:border-slate-700 dark:border-t-violet-400"></div>
-          <span class="text-sm font-semibold text-slate-400 dark:text-slate-500">加载中...</span>
+        <div v-if="loading && !records.length" class="flex items-center justify-center py-10">
+          <div class="h-5 w-5 animate-spin rounded-full border-2 border-white/10 border-t-[rgb(129,115,223)]"></div>
         </div>
 
         <!-- 空状态 -->
-        <div v-else-if="!loading && !records.length" class="flex flex-col items-center justify-center py-20">
-          <div class="mb-4 flex h-20 w-20 items-center justify-center rounded-3xl bg-slate-100 dark:bg-white/5">
-            <i class="fa-solid fa-clock-rotate-left text-3xl text-slate-300 dark:text-slate-600"></i>
-          </div>
-          <p class="text-base font-bold text-slate-400 dark:text-slate-500">暂无分割记录</p>
-          <p class="mt-1 text-xs text-slate-400 dark:text-slate-600">完成题目分割后，记录将自动保存在这里</p>
+        <div v-else-if="!loading && !records.length" class="px-4 py-10 text-center">
+          <i class="fa-solid fa-clock-rotate-left text-lg text-[#62666d] mb-2"></i>
+          <p class="text-xs text-[#62666d]">暂无记录</p>
         </div>
 
         <!-- 记录列表 -->
-        <div v-else class="space-y-3">
+        <div v-else>
           <div
             v-for="(r, idx) in records"
             :key="r.id"
-            class="group cursor-pointer rounded-2xl border transition-all duration-300"
-            :class="activeRecord?.id === r.id
-              ? 'border-violet-400/60 bg-violet-50/50 shadow-lg shadow-violet-500/10 dark:border-violet-500/40 dark:bg-violet-500/[0.06] dark:shadow-violet-500/5'
-              : 'border-slate-200/60 bg-white/70 hover:border-violet-300/50 hover:bg-white/90 hover:shadow-md dark:border-white/[0.06] dark:bg-white/[0.03] dark:hover:border-violet-500/20 dark:hover:bg-white/[0.05]'"
+            class="group cursor-pointer border-b border-white/[0.05] px-4 py-3 transition-colors hover:bg-white/[0.03]"
+            :class="activeRecord?.id === r.id ? 'bg-white/[0.04]' : ''"
             @click="openDetail(r)"
           >
-            <!-- 记录卡片头 -->
-            <div class="flex items-center gap-4 px-5 py-4">
-              <!-- 序号圆 -->
-              <div
-                class="flex h-10 w-10 shrink-0 items-center justify-center rounded-xl text-sm font-black transition-colors"
-                :class="activeRecord?.id === r.id
-                  ? 'bg-violet-500 text-white shadow-md shadow-violet-500/30'
-                  : 'bg-slate-100 text-slate-500 group-hover:bg-violet-100 group-hover:text-violet-600 dark:bg-white/[0.06] dark:text-slate-400 dark:group-hover:bg-violet-500/20 dark:group-hover:text-violet-300'"
-              >
-                {{ records.length - idx }}
-              </div>
-
-              <!-- 主信息 -->
-              <div class="min-w-0 flex-1">
-                <div class="flex items-center gap-2 mb-1">
-                  <span class="truncate text-sm font-bold text-slate-800 dark:text-white">
-                    {{ r.subject || '未识别科目' }}
-                  </span>
-                  <span class="rounded-full px-2 py-0.5 text-[10px] font-bold" :class="providerColor(r.model_provider)">
-                    {{ providerLabel(r.model_provider) }}
-                  </span>
-                </div>
-                <div class="flex items-center gap-3 text-xs text-slate-400 dark:text-slate-500">
-                  <span class="inline-flex items-center gap-1">
-                    <i class="fa-regular fa-file text-[10px]"></i>
-                    {{ (r.file_names || []).length }} 个文件
-                  </span>
-                  <span class="h-3 w-px bg-slate-200 dark:bg-white/10"></span>
-                  <span class="inline-flex items-center gap-1" :title="formatFullDate(r.created_at)">
-                    <i class="fa-regular fa-clock text-[10px]"></i>
-                    {{ formatDate(r.created_at) }}
-                  </span>
-                </div>
-              </div>
-
-              <!-- 题目数 + 展开箭头 -->
-              <div class="flex items-center gap-3 shrink-0">
-                <div class="text-right">
-                  <div class="text-xl font-black tabular-nums text-slate-800 dark:text-white">{{ r.question_count }}</div>
-                  <div class="text-[10px] font-bold text-slate-400 dark:text-slate-500">道题目</div>
-                </div>
-                <i
-                  class="fa-solid fa-chevron-down text-xs text-slate-300 transition-transform duration-300 dark:text-slate-600"
-                  :class="{ '-rotate-180': activeRecord?.id === r.id }"
-                ></i>
-              </div>
+            <!-- 一行：科目 + 时间 + 题数 -->
+            <div class="flex items-center justify-between mb-1">
+              <span class="text-sm font-medium text-[#f7f8f8] truncate">{{ r.subject || '未识别' }}</span>
+              <span class="text-xs tabular-nums text-[rgb(129,115,223)]">{{ r.question_count }} 题</span>
             </div>
-
-            <!-- 文件名列表 -->
-            <div v-if="r.file_names?.length" class="flex flex-wrap gap-1.5 px-5 pb-3">
-              <span
-                v-for="fname in r.file_names"
-                :key="fname"
-                class="inline-flex items-center gap-1 rounded-lg bg-slate-100/80 px-2 py-0.5 text-[11px] font-semibold text-slate-500 dark:bg-white/[0.04] dark:text-slate-400"
-              >
-                <i class="fa-solid text-[9px]" :class="fname.toLowerCase().endsWith('.pdf') ? 'fa-file-pdf text-rose-400' : 'fa-file-image text-blue-400'"></i>
-                {{ fname }}
-              </span>
+            <!-- 二行：文件数 + 时间 -->
+            <div class="flex items-center gap-2 text-xs text-[#62666d]">
+              <span>{{ (r.file_names || []).length }} 个文件</span>
+              <span>·</span>
+              <span :title="formatFullDate(r.created_at)">{{ formatDate(r.created_at) }}</span>
             </div>
 
             <!-- 展开详情面板 -->
@@ -318,10 +243,10 @@ defineExpose({ refresh: loadRecords })
                   <div class="grid grid-cols-1 gap-2.5 lg:grid-cols-2">
                     <div
                       v-for="q in activeQuestions"
-                      :key="q.question_id"
-                      @click.stop="toggleSelect(q.question_id)"
+                      :key="q.uid"
+                      @click.stop="toggleSelect(q.uid)"
                       class="cursor-pointer rounded-xl border p-3.5 transition-all duration-200"
-                      :class="selectedIds.has(q.question_id)
+                      :class="selectedIds.has(q.uid)
                         ? 'border-violet-400/60 bg-violet-50/60 ring-1 ring-violet-400/30 dark:border-violet-500/40 dark:bg-violet-500/[0.08] dark:ring-violet-500/20'
                         : 'border-slate-200/50 bg-white/50 hover:border-violet-300/40 hover:bg-white/80 dark:border-white/[0.05] dark:bg-white/[0.02] dark:hover:border-violet-500/20'"
                     >
@@ -329,11 +254,11 @@ defineExpose({ refresh: loadRecords })
                         <!-- 选中指示 -->
                         <div
                           class="flex h-5 w-5 shrink-0 items-center justify-center rounded-md border text-[10px] transition-colors"
-                          :class="selectedIds.has(q.question_id)
+                          :class="selectedIds.has(q.uid)
                             ? 'border-violet-500 bg-violet-500 text-white dark:border-violet-400 dark:bg-violet-500'
                             : 'border-slate-300 dark:border-white/15'"
                         >
-                          <i v-if="selectedIds.has(q.question_id)" class="fa-solid fa-check"></i>
+                          <i v-if="selectedIds.has(q.uid)" class="fa-solid fa-check"></i>
                         </div>
                         <span class="text-xs font-black text-slate-700 dark:text-slate-200">
                           {{ q.question_id || '?' }}
diff --git a/frontend/src/views/app/WorkspaceView.vue b/frontend/src/views/app/WorkspaceView.vue
new file mode 100644
index 00000000..ce82f57c
--- /dev/null
+++ b/frontend/src/views/app/WorkspaceView.vue
@@ -0,0 +1,242 @@
+<script setup>
+/**
+ * WorkspaceView.vue
+ * 录入工作台 — 上传/擦除/OCR/分割/导出 全流程
+ */
+import { computed, nextTick, onBeforeUnmount, ref, watch } from 'vue'
+import { useToast } from '@/composables/useToast.js'
+import { useSystemStatus } from '@/composables/useSystemStatus.js'
+import { useImageModal } from '@/composables/useImageModal.js'
+import { useWorkspaceNav } from '@/composables/useWorkspaceNav.js'
+import { useQuestionList } from '@/composables/useQuestionList.js'
+import { useFileUpload } from '@/composables/useFileUpload.js'
+import { useSplitPipeline } from '@/composables/useSplitPipeline.js'
+import ContentPanel from '@/components/workspace/ContentPanel.vue'
+import SelectionPanel from '@/components/workspace/SelectionPanel.vue'
+import UploadStage from '@/components/workspace/UploadStage.vue'
+import ReviewStage from '@/components/workspace/ReviewStage.vue'
+import SplitHistory from '@/views/app/SplitHistoryView.vue'
+
+const { pushToast } = useToast()
+const { openModal } = useImageModal()
+const { currentView } = useWorkspaceNav()
+const {
+  statusLoading, statusError, selectedModel, selectedProvider,
+  providerOptions, hasConfiguredModel, statusPills,
+} = useSystemStatus()
+const {
+  questions, selectedIds,
+  toggleQuestion, selectAll, deselectAll, reorderQuestions, typesetMath,
+} = useQuestionList()
+
+const props = defineProps({
+  theme: { type: String, default: 'dark' },
+})
+
+// ── 步骤控制 ────────────────────────────────────────────
+const step = ref(1)
+const uploadMode = ref('exam')
+const splitting = ref(false)
+const splitCompleted = ref(false)
+const showSplitHistory = ref(false)
+
+const S = computed(() => {
+  const off = eraseEnabled.value ? 1 : 0
+  return { UPLOAD: 1, ERASE: 2, OCR: 2 + off, SPLIT: 3 + off, EXPORT: 4 + off }
+})
+const workspaceSteps = computed(() => {
+  const labels = uploadMode.value === 'note'
+    ? (eraseEnabled.value ? ['上传', '擦除', 'OCR', '整理', '保存'] : ['上传', 'OCR', '整理', '保存'])
+    : (eraseEnabled.value ? ['上传', '擦除', 'OCR', '分割', '导出'] : ['上传', 'OCR', '分割', '导出'])
+  return labels.map((label, i) => ({
+    label,
+    done: i + 1 < step.value,
+    active: i + 1 === step.value,
+  }))
+})
+
+// ── 文件上传 ────────────────────────────────────────────
+const {
+  uploadBusy, uploadReady, pendingFiles, pendingPreviewUrls,
+  fileProgress, waitingKeys,
+  enqueueUpload, removePendingFile, stopFakeProgress, doReset: _doReset,
+} = useFileUpload(pushToast, S, questions, selectedIds, splitCompleted, splitting, uploadMode)
+
+const splitEnabled = computed(() => !splitting.value && !splitCompleted.value && uploadReady.value && !uploadBusy.value && hasConfiguredModel.value)
+
+const doReset = () => {
+  _doReset(providerOptions, selectedModel, step)
+  step.value = S.value.UPLOAD
+}
+
+// ── 分割流水线 ──────────────────────────────────────────
+const {
+  eraseEnabled, eraseLoading, eraseImages, eraseDone,
+  ocrLoading, ocrPages, ocrDone,
+  startProcess, doErase, doOcr, doSplit, doExport, doSaveToDb,
+} = useSplitPipeline(pushToast, currentView, step, S, uploadReady, splitting, splitCompleted, uploadMode, selectedProvider, selectedModel, questions, selectedIds, pendingFiles, typesetMath)
+
+const reviewStageRef = ref(null)
+
+const handleLoadRecord = (qs, record) => {
+  questions.value = qs || []; selectedIds.clear()
+  splitCompleted.value = true; step.value = S.value.EXPORT
+  currentView.value = 'workspace_review'
+  pushToast('success', `已加载「${record?.subject || '历史记录'}」的 ${qs.length} 道题目`)
+  nextTick(() => typesetMath())
+}
+
+const handleBack = () => {
+  doReset()
+  eraseImages.value = []; eraseDone.value = false
+  ocrPages.value = []; ocrDone.value = false
+  currentView.value = 'workspace'
+}
+
+// ── 键盘事件 ────────────────────────────────────────────
+const onKeydown = (e) => {
+  if (e.key === 'a' && (e.ctrlKey || e.metaKey) && questions.value.length) {
+    e.preventDefault(); selectAll()
+  }
+}
+
+// ── 视图切换 ────────────────────────────────────────────
+watch(currentView, async (v) => {
+  if (v === 'workspace_review') {
+    await nextTick()
+    setTimeout(() => { reviewStageRef.value?.triggerTypeset?.() }, 650)
+  }
+})
+
+// ── 生命周期 ────────────────────────────────────────────
+import { onMounted } from 'vue'
+onMounted(() => document.addEventListener('keydown', onKeydown))
+onBeforeUnmount(() => {
+  stopFakeProgress()
+  document.removeEventListener('keydown', onKeydown)
+})
+</script>
+
+<template>
+  <div class="h-full">
+    <Transition name="flip" mode="out-in">
+      <!-- 第一页：上传与分析 -->
+      <ContentPanel v-if="currentView === 'workspace'" key="upload" title="智能录入与分析" :steps="workspaceSteps" :current-step="step - 1">
+        <template #toolbar>
+          <button
+            @click="showSplitHistory = !showSplitHistory"
+            class="inline-flex items-center gap-2 rounded-md px-3 py-1.5 text-xs font-medium transition-colors"
+            :class="showSplitHistory
+              ? 'bg-white/[0.06] text-[#f7f8f8] border border-white/[0.12]'
+              : 'border border-white/[0.08] bg-white/[0.02] text-[#d0d6e0] hover:bg-white/[0.05] hover:border-white/[0.12]'"
+          >
+            <i class="fa-solid fa-clock-rotate-left text-[10px]"></i>
+            分割历史
+          </button>
+        </template>
+
+        <template v-if="showSplitHistory" #sidebar>
+          <SplitHistory
+            :theme="theme"
+            :visible="showSplitHistory"
+            @push-toast="pushToast"
+            @open-image="openModal"
+            @load-record="(r) => { handleLoadRecord(r); showSplitHistory = false }"
+            @go-workspace="currentView = splitCompleted ? 'workspace_review' : 'workspace'"
+          />
+        </template>
+
+        <UploadStage
+          :upload-mode="uploadMode"
+          :erase-enabled="eraseEnabled"
+          :status-loading="statusLoading"
+          :status-error="statusError"
+          :status-pills="statusPills"
+          :provider-options="providerOptions"
+          :selected-model="selectedModel"
+          :has-configured-model="hasConfiguredModel"
+          :splitting="splitting"
+          :split-completed="splitCompleted"
+          :pending-files="pendingFiles"
+          :file-progress="fileProgress"
+          :waiting-keys="waitingKeys"
+          :upload-busy="uploadBusy"
+          :upload-ready="uploadReady"
+          :split-enabled="splitEnabled"
+          @update:upload-mode="(v) => uploadMode = v"
+          @update:erase-enabled="(v) => eraseEnabled = v"
+          @update:selected-model="(v) => selectedModel = v"
+          @upload="enqueueUpload"
+          @remove-file="removePendingFile"
+          @split="startProcess"
+        />
+      </ContentPanel>
+
+      <!-- 第二页：解析结果核对 -->
+      <ContentPanel
+        v-else-if="currentView === 'workspace_review'"
+        key="review"
+        :title="eraseLoading ? '正在擦除...' : eraseDone && !ocrLoading && !ocrDone ? '擦除预览' : splitting ? '正在分割...' : ocrDone && !splitCompleted ? 'OCR 预览' : '题目数据核对'"
+        :steps="workspaceSteps"
+        :current-step="step - 1"
+      >
+        <template #toolbar>
+          <button
+            @click="handleBack"
+            class="group inline-flex items-center gap-2 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#d0d6e0] hover:bg-white/[0.05] hover:border-white/[0.12] transition-colors"
+          >
+            <i class="fa-solid fa-arrow-left-long text-xs transition-transform group-hover:-translate-x-0.5"></i>
+            返回
+          </button>
+          <template v-if="eraseDone && !ocrLoading && !ocrDone">
+            <button @click="doErase" class="inline-flex items-center gap-1.5 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#d0d6e0] hover:bg-white/[0.05] transition-colors">
+              <i class="fa-solid fa-arrows-rotate text-[10px]"></i> 重新擦除
+            </button>
+            <button @click="doOcr" class="inline-flex items-center gap-1.5 rounded-md bg-[rgb(129,115,223)] px-3 py-1.5 text-xs font-medium text-white hover:bg-[rgb(145,132,235)] transition-colors">
+              <i class="fa-solid fa-check text-[10px]"></i> 确认，开始 OCR
+            </button>
+          </template>
+          <template v-else-if="ocrDone && !splitCompleted && !splitting">
+            <button @click="doOcr" class="inline-flex items-center gap-1.5 rounded-md border border-white/[0.08] bg-white/[0.02] px-3 py-1.5 text-xs font-medium text-[#d0d6e0] hover:bg-white/[0.05] transition-colors">
+              <i class="fa-solid fa-arrows-rotate text-[10px]"></i> 重新识别
+            </button>
+            <button @click="doSplit" class="inline-flex items-center gap-1.5 rounded-md bg-[rgb(129,115,223)] px-3 py-1.5 text-xs font-medium text-white hover:bg-[rgb(145,132,235)] transition-colors">
+              <i class="fa-solid fa-check text-[10px]"></i> 确认并分割
+            </button>
+          </template>
+        </template>
+
+        <ReviewStage
+          ref="reviewStageRef"
+          :erase-loading="eraseLoading"
+          :erase-done="eraseDone"
+          :erase-images="eraseImages"
+          :ocr-loading="ocrLoading"
+          :ocr-done="ocrDone"
+          :ocr-pages="ocrPages"
+          :splitting="splitting"
+          :split-completed="splitCompleted"
+          :questions="questions"
+          :selected-ids="selectedIds"
+          :preview-url="pendingPreviewUrls[0]"
+          @toggle-select="toggleQuestion"
+          @select-all="selectAll"
+          @deselect-all="deselectAll"
+          @open-image="openModal"
+          @reorder="reorderQuestions"
+        />
+      </ContentPanel>
+    </Transition>
+
+    <!-- 浮动选择面板 -->
+    <SelectionPanel
+      :visible="currentView === 'workspace_review'"
+      :count="selectedIds.size"
+      export-label="导出错题本"
+      :show-save="true"
+      @export="doExport"
+      @save="doSaveToDb"
+      @clear="deselectAll"
+    />
+  </div>
+</template>
diff --git a/frontend/src/views/auth/AuthLayout.vue b/frontend/src/views/auth/AuthLayout.vue
new file mode 100644
index 00000000..7a2b5d94
--- /dev/null
+++ b/frontend/src/views/auth/AuthLayout.vue
@@ -0,0 +1,286 @@
+<script setup>
+import { ref, onMounted, onUnmounted } from 'vue'
+import { useRoute, useRouter } from 'vue-router'
+import { useTheme } from '@/composables/useTheme.js'
+import HomePill from '@/components/home/HomePill.vue'
+import BaseLogo from '@/components/base/BaseLogo.vue'
+
+const { initTheme } = useTheme()
+const route = useRoute()
+const router = useRouter()
+const transitionName = ref('auth-slide-left')
+
+onMounted(() => {
+  initTheme()
+})
+
+router.beforeEach((to, from) => {
+  const toOrder = to.meta.order ?? 0
+  const fromOrder = from.meta.order ?? 0
+  transitionName.value = toOrder >= fromOrder ? 'auth-slide-left' : 'auth-slide-right'
+})
+
+// ── 鼠标跟随发光 ──
+const brandRef = ref(null)
+const featureRefs = ref([])
+
+function handleMouseMove(e) {
+  const el = brandRef.value
+  if (!el) return
+  // 面板级坐标（用于曲线发光）
+  const rect = el.getBoundingClientRect()
+  el.style.setProperty('--mx', `${e.clientX - rect.left}px`)
+  el.style.setProperty('--my', `${e.clientY - rect.top}px`)
+  // 每个图标的相对坐标
+  featureRefs.value.forEach(item => {
+    if (!item) return
+    const r = item.getBoundingClientRect()
+    item.style.setProperty('--ix', `${e.clientX - r.left}px`)
+    item.style.setProperty('--iy', `${e.clientY - r.top}px`)
+  })
+}
+
+onMounted(() => {
+  brandRef.value?.addEventListener('mousemove', handleMouseMove)
+})
+onUnmounted(() => {
+  brandRef.value?.removeEventListener('mousemove', handleMouseMove)
+})
+
+const FEATURES = [
+  { icon: 'fa-camera', text: '拍照上传，AI 自动识别题目与公式' },
+  { icon: 'fa-brain', text: 'LangGraph Agent 智能拆题纠错' },
+  { icon: 'fa-tags', text: '知识点自动打标，构建个人图谱' },
+  { icon: 'fa-file-export', text: '一键导出 Markdown / PDF 错题本' },
+]
+
+// ── 随机星星 ──
+const stars = ref([])
+
+onMounted(() => {
+  const list = []
+  for (let i = 0; i < 30; i++) {
+    list.push({
+      left: Math.random() * 100,
+      top: 55 + Math.random() * 42,
+      size: 1 + Math.random() * 2,
+      opacity: 0.08 + Math.random() * 0.35,
+      duration: 2 + Math.random() * 3,  // 2~5s
+      delay: Math.random() * 4,          // 0~4s 延迟
+    })
+  }
+  stars.value = list
+})
+</script>
+
+<template>
+  <div class="min-h-screen flex bg-[#0A0A0F]">
+
+    <!-- 左侧品牌区（大屏显示） -->
+    <div ref="brandRef" class="hidden lg:flex flex-col justify-between w-[52%] relative overflow-hidden p-12 bg-gradient-to-br from-[#12121a] to-[#0A0A0F]">
+      <!-- 底层：紫色曲线（暗态） -->
+      <svg class="absolute inset-0 w-full h-full pointer-events-none" viewBox="0 0 600 800" preserveAspectRatio="none" fill="none">
+        <path d="M-50,250 C120,400 280,100 420,280 C560,460 600,180 700,350" stroke="rgba(129,115,223,0.12)" stroke-width="2.5" />
+        <path d="M-80,500 C80,650 240,340 400,520 C560,700 620,400 720,580" stroke="rgba(129,115,223,0.08)" stroke-width="2" />
+      </svg>
+      <!-- 亮层：鼠标跟随发光的紫色曲线（通过 mask 只显示鼠标附近） -->
+      <div class="auth-glow-layer absolute inset-0 pointer-events-none">
+        <svg class="w-full h-full" viewBox="0 0 600 800" preserveAspectRatio="none" fill="none">
+          <path d="M-50,250 C120,400 280,100 420,280 C560,460 600,180 700,350" stroke="rgba(151,137,222,0.9)" stroke-width="3" style="filter: drop-shadow(0 0 14px rgba(129,115,223,0.8)) drop-shadow(0 0 40px rgba(129,115,223,0.4));" />
+          <path d="M-80,500 C80,650 240,340 400,520 C560,700 620,400 720,580" stroke="rgba(151,137,222,0.7)" stroke-width="2.5" style="filter: drop-shadow(0 0 12px rgba(129,115,223,0.7)) drop-shadow(0 0 30px rgba(129,115,223,0.3));" />
+        </svg>
+      </div>
+      <!-- 底部紫色光晕 -->
+      <div class="absolute bottom-0 left-1/2 -translate-x-1/2 w-[80%] h-[200px] pointer-events-none rounded-full bg-indigo-500/[0.08] blur-[80px]"></div>
+
+      <!-- 不规则星星 -->
+      <div class="absolute inset-0 pointer-events-none">
+        <div
+          v-for="(s, i) in stars"
+          :key="i"
+          class="absolute rounded-full bg-white star-twinkle"
+          :style="{
+            left: s.left + '%',
+            top: s.top + '%',
+            width: s.size + 'px',
+            height: s.size + 'px',
+            '--star-opacity': s.opacity,
+            animationDuration: s.duration + 's',
+            animationDelay: s.delay + 's',
+          }"
+        ></div>
+      </div>
+
+      <!-- 右侧分割线 -->
+      <div class="absolute top-0 right-0 bottom-0 w-px pointer-events-none bg-gradient-to-b from-transparent via-white/[0.08] to-transparent"></div>
+
+      <!-- 顶部 Logo -->
+      <div class="relative flex items-center gap-3">
+        <BaseLogo breathe />
+        <span class="text-base font-semibold text-white/80 tracking-wide">智卷错题本</span>
+      </div>
+
+      <!-- 中部主文案 -->
+      <div class="relative">
+        <HomePill class="mb-6" />
+        <h2 class="text-4xl font-semibold text-white leading-tight mb-4 tracking-tight">
+          重塑错题整理<br />
+          <span class="text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">一键生成知识图谱</span>
+        </h2>
+        <p class="text-sm text-white/45 leading-relaxed max-w-sm">
+          专为中学生与大学生研发。上传凌乱试卷，AI 自动完成图片分割、OCR 纠错及 LaTeX 公式还原。
+        </p>
+
+        <!-- 特性列表 — 鼠标跟随染色图标 -->
+        <ul class="mt-8 space-y-4">
+          <li v-for="(f, i) in FEATURES" :key="f.text" :ref="el => featureRefs[i] = el" class="flex items-center gap-3 text-sm text-white/60">
+            <!-- 图标容器: 双层结构 -->
+            <div class="relative flex h-8 w-8 shrink-0 items-center justify-center rounded-xl p-px overflow-hidden">
+              <!-- 默认边框 -->
+              <div class="absolute inset-0 bg-white/[0.08] rounded-xl"></div>
+              <!-- 鼠标跟随边框高光 -->
+              <div class="pointer-events-none absolute inset-0"
+                style="background: radial-gradient(80px circle at var(--ix, -500px) var(--iy, -500px), rgba(151,137,222,0.7), transparent 50%);"></div>
+              <!-- 图标内部 -->
+              <div class="relative h-full w-full bg-[#15151e] rounded-[11px] flex items-center justify-center">
+                <!-- 白色底层图标 -->
+                <i :class="`fas ${f.icon} text-xs text-white/50 absolute`"></i>
+                <!-- 鼠标跟随染色图标 -->
+                <div class="absolute inset-0 flex items-center justify-center"
+                  style="mask-image: radial-gradient(80px circle at var(--ix, -500px) var(--iy, -500px), black 0%, transparent 100%);
+                         -webkit-mask-image: radial-gradient(80px circle at var(--ix, -500px) var(--iy, -500px), black 0%, transparent 100%);">
+                  <i :class="`fas ${f.icon} text-xs text-[#9789de] drop-shadow-[0_0_8px_rgba(151,137,222,0.8)]`"></i>
+                </div>
+              </div>
+            </div>
+            {{ f.text }}
+          </li>
+        </ul>
+      </div>
+
+      <!-- 底部版权 -->
+      <div class="relative text-xs text-white/20">
+        © {{ new Date().getFullYear() }} 智卷错题本 · All rights reserved
+      </div>
+    </div>
+
+    <!-- 右侧表单区 -->
+    <div class="flex-1 flex flex-col items-center justify-center px-4 py-12 lg:px-16 relative bg-[#0e0e16]">
+
+      <!-- 返回主页 -->
+      <a href="/" class="absolute top-6 right-6 inline-flex items-center gap-2 px-4 py-2 rounded-lg text-xs font-medium text-white/50 bg-white/[0.03] border border-white/[0.06] hover:bg-white/[0.06] hover:text-white/70 transition-all">
+        <i class="fas fa-arrow-left text-xs"></i>
+        返回主页
+      </a>
+
+      <!-- 移动端 Logo（小屏显示） -->
+      <div class="lg:hidden text-center mb-8">
+        <div class="relative inline-flex mb-4">
+          <div class="bg-white/[0.04] border border-white/[0.08] p-2.5 rounded-xl">
+            <img src="/logo.svg" class="w-7 h-7 brightness-0 invert opacity-70" alt="logo" />
+          </div>
+        </div>
+        <h1 class="text-2xl font-semibold text-white tracking-wide">智卷错题本</h1>
+      </div>
+
+      <div class="w-full max-w-sm">
+        <!-- 标题 -->
+        <div class="mb-8">
+          <h3 class="text-2xl font-semibold text-white">
+            {{ route.path === '/auth/login' ? '欢迎回来' : '创建账户' }}
+          </h3>
+          <p class="text-sm text-white/35 mt-1">
+            {{ route.path === '/auth/login' ? '登录以继续使用你的错题本' : '免费注册，开始智能错题整理' }}
+          </p>
+        </div>
+
+        <!-- Tab 切换 -->
+        <div class="flex rounded-xl bg-white/[0.03] border border-white/[0.06] p-1 mb-6">
+          <RouterLink
+            to="/auth/login"
+            class="flex-1 py-2 text-sm font-medium rounded-lg text-center transition-all"
+            :class="route.path === '/auth/login'
+              ? 'bg-white/[0.08] text-white'
+              : 'text-white/35 hover:text-white/60'"
+          >登录</RouterLink>
+          <RouterLink
+            to="/auth/register"
+            class="flex-1 py-2 text-sm font-medium rounded-lg text-center transition-all"
+            :class="route.path === '/auth/register'
+              ? 'bg-white/[0.08] text-white'
+              : 'text-white/35 hover:text-white/60'"
+          >注册</RouterLink>
+        </div>
+
+        <!-- 表单内容（滑动过渡） -->
+        <RouterView v-slot="{ Component }">
+          <Transition :name="transitionName" mode="out-in">
+            <component :is="Component" :key="route.path" />
+          </Transition>
+        </RouterView>
+
+        <!-- 底部版权（移动端） -->
+        <p class="lg:hidden text-center text-xs text-white/20 mt-8">
+          © {{ new Date().getFullYear() }} 智卷错题本
+        </p>
+      </div>
+    </div>
+
+  </div>
+</template>
+
+<style>
+/* 禁用浏览器自动填充背景色 */
+input:-webkit-autofill,
+input:-webkit-autofill:hover,
+input:-webkit-autofill:focus {
+  transition: background-color 9999s ease;
+  -webkit-text-fill-color: #e2e8f0;
+  caret-color: #e2e8f0;
+}
+</style>
+
+<style scoped>
+/* 星星闪烁 */
+@keyframes star-twinkle {
+  0%, 100% { opacity: var(--star-opacity); }
+  50% { opacity: 0.02; }
+}
+.star-twinkle {
+  animation: star-twinkle ease-in-out infinite;
+}
+
+/* 鼠标跟随发光层 */
+.auth-glow-layer {
+  mask-image: radial-gradient(350px circle at var(--mx, -500px) var(--my, -500px), black 0%, transparent 100%);
+  -webkit-mask-image: radial-gradient(350px circle at var(--mx, -500px) var(--my, -500px), black 0%, transparent 100%);
+}
+
+/* 紫-白-紫渐变扫光 */
+@keyframes gradient-sweep {
+  0% { background-position: -100% center; }
+  100% { background-position: 100% center; }
+}
+.animate-gradient-sweep {
+  animation: gradient-sweep 3s ease-in-out infinite alternate;
+}
+
+/* 向左滑（登录 → 注册） */
+.auth-slide-left-enter-active,
+.auth-slide-left-leave-active {
+  transition: opacity 0.22s ease, transform 0.22s ease;
+}
+.auth-slide-left-enter-from  { opacity: 0; transform: translateX(32px); }
+.auth-slide-left-leave-to   { opacity: 0; transform: translateX(-32px); }
+
+/* 向右滑（注册 → 登录） */
+.auth-slide-right-enter-active,
+.auth-slide-right-leave-active {
+  transition: opacity 0.22s ease, transform 0.22s ease;
+}
+.auth-slide-right-enter-from { opacity: 0; transform: translateX(-32px); }
+.auth-slide-right-leave-to  { opacity: 0; transform: translateX(32px); }
+</style>
diff --git a/frontend/src/views/auth/LoginView.vue b/frontend/src/views/auth/LoginView.vue
new file mode 100644
index 00000000..ad4a42dd
--- /dev/null
+++ b/frontend/src/views/auth/LoginView.vue
@@ -0,0 +1,81 @@
+<script setup>
+import { ref, reactive } from 'vue'
+import { useRouter } from 'vue-router'
+import { useAuth } from '@/composables/useAuth.js'
+import BaseButton from '@/components/base/BaseButton.vue'
+import ForgotPasswordModal from '@/components/auth/ForgotPasswordModal.vue'
+
+const router = useRouter()
+const { currentUser } = useAuth()
+
+const loading = ref(false)
+const showPwd = ref(false)
+const error = ref('')
+const form = reactive({ email: '', password: '' })
+const fpOpen = ref(false)
+
+async function handleLogin() {
+  error.value = ''
+  loading.value = true
+  try {
+    const res = await fetch('/api/auth/login', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ email: form.email, password: form.password }),
+    })
+    const data = await res.json()
+    if (!res.ok) {
+      error.value = data.error || '登录失败'
+    } else {
+      currentUser.value = data.user
+      router.push('/app')
+    }
+  } catch {
+    error.value = '网络错误，请重试'
+  } finally {
+    loading.value = false
+  }
+}
+</script>
+
+<template>
+  <div>
+  <form @submit.prevent="handleLogin" class="space-y-4">
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">邮箱或用户名</label>
+      <input v-model="form.email" type="text" required autocomplete="username" placeholder="邮箱 或 用户名"
+        class="w-full h-10 px-4 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm" />
+    </div>
+
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">密码</label>
+      <div class="relative">
+        <input v-model="form.password" :type="showPwd ? 'text' : 'password'" required autocomplete="current-password"
+          placeholder="请输入密码"
+          class="w-full h-10 px-4 pr-11 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm" />
+        <button type="button" @click="showPwd = !showPwd"
+          class="absolute right-3 top-1/2 -translate-y-1/2 text-white/25 hover:text-white/50 transition-colors">
+          <i :class="showPwd ? 'fas fa-eye-slash' : 'fas fa-eye'" class="text-xs"></i>
+        </button>
+      </div>
+    </div>
+    <div class="flex justify-end">
+      <button type="button" @click="fpOpen = true"
+        class="text-xs text-white/40 hover:text-indigo-400 transition-colors">
+        忘记密码？
+      </button>
+    </div>
+    <p v-if="error" class="text-sm text-rose-400 flex items-center gap-2">
+      <i class="fas fa-circle-exclamation text-xs"></i>{{ error }}
+    </p>
+
+
+    <BaseButton variant="cta" type="submit" class="w-full" :disabled="loading">
+      <i v-if="loading" class="fas fa-spinner fa-spin text-xs"></i>
+      <span>{{ loading ? '登录中...' : '登录' }}</span>
+    </BaseButton>
+  </form>
+
+  <ForgotPasswordModal :open="fpOpen" @close="fpOpen = false" />
+  </div>
+</template>
diff --git a/frontend/src/views/auth/RegisterView.vue b/frontend/src/views/auth/RegisterView.vue
new file mode 100644
index 00000000..82fb490b
--- /dev/null
+++ b/frontend/src/views/auth/RegisterView.vue
@@ -0,0 +1,255 @@
+<script setup>
+import { ref, reactive, onBeforeUnmount } from 'vue'
+import { useRouter } from 'vue-router'
+import { useAuth } from '@/composables/useAuth.js'
+import BaseButton from '@/components/base/BaseButton.vue'
+
+const router = useRouter()
+const { currentUser } = useAuth()
+
+const loading = ref(false)
+const showPwd = ref(false)
+const error = ref('')
+const success = ref('')
+const form = reactive({ username: '', email: '', password: '', confirm: '', code: '' })
+
+const sendingCode = ref(false)
+const countdown = ref(0)
+const rateLimitCountdown = ref(0)
+let countdownTimer = null
+let rateLimitTimer = null
+let sendDebounceTimer = null
+
+const passwordMismatch = () => form.confirm && form.password !== form.confirm
+
+function clearCountdown() {
+  if (countdownTimer) {
+    clearInterval(countdownTimer)
+    countdownTimer = null
+  }
+}
+
+function startCountdown(seconds = 60) {
+  clearCountdown()
+  countdown.value = seconds
+  countdownTimer = setInterval(() => {
+    countdown.value -= 1
+    if (countdown.value <= 0) {
+      clearCountdown()
+      countdown.value = 0
+    }
+  }, 1000)
+}
+
+function clearRateLimitCountdown() {
+  if (rateLimitTimer) {
+    clearInterval(rateLimitTimer)
+    rateLimitTimer = null
+  }
+}
+
+function startRateLimitCountdown(seconds) {
+  clearRateLimitCountdown()
+  rateLimitCountdown.value = seconds
+  error.value = `发送过于频繁，请 ${seconds} 秒后再试`
+  rateLimitTimer = setInterval(() => {
+    rateLimitCountdown.value -= 1
+    if (rateLimitCountdown.value <= 0) {
+      clearRateLimitCountdown()
+      rateLimitCountdown.value = 0
+      error.value = ''
+    } else {
+      error.value = `发送过于频繁，请 ${rateLimitCountdown.value} 秒后再试`
+    }
+  }, 1000)
+}
+
+async function handleSendCode() {
+  error.value = ''
+  clearRateLimitCountdown()
+  if (sendingCode.value || countdown.value > 0) return
+  const email = (form.email || '').trim()
+  if (!email || !email.includes('@')) {
+    error.value = '请先填写有效邮箱'
+    return
+  }
+  if (sendDebounceTimer) clearTimeout(sendDebounceTimer)
+  sendDebounceTimer = setTimeout(async () => {
+    sendDebounceTimer = null
+    sendingCode.value = true
+    try {
+      const res = await fetch('/api/auth/send-code', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        credentials: 'include',
+        body: JSON.stringify({ email, type: 'register' }),
+      })
+      const data = await res.json().catch(() => ({}))
+      if (!res.ok) {
+        // 429 频率限制：解析剩余秒数，启动错误提示倒计时
+        if (res.status === 429) {
+          const match = (data.error || '').match(/(\d+)\s*秒/)
+          const wait = match ? parseInt(match[1], 10) : 30
+          startRateLimitCountdown(wait)
+        } else {
+          error.value = data.error || '发送失败'
+        }
+        return
+      }
+      success.value = '验证码已发送到邮箱，请查收'
+      startCountdown(60)
+    } catch {
+      error.value = '网络错误，请重试'
+    } finally {
+      sendingCode.value = false
+    }
+  }, 280)
+}
+
+async function handleRegister() {
+  error.value = ''
+  success.value = ''
+  if (passwordMismatch()) { error.value = '两次密码不一致'; return }
+  if (!form.code.trim()) { error.value = '请输入验证码'; return }
+  loading.value = true
+  try {
+    const res = await fetch('/api/auth/register', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      credentials: 'include',
+      body: JSON.stringify({
+        username: form.username,
+        email: form.email,
+        password: form.password,
+        code: form.code.trim(),
+      }),
+    })
+    const data = await res.json()
+    if (!res.ok) {
+      error.value = data.error || '注册失败'
+    } else {
+      success.value = '注册成功！正在登录...'
+      currentUser.value = data.user
+      router.push('/app')
+    }
+  } catch {
+    error.value = '网络错误，请重试'
+  } finally {
+    loading.value = false
+  }
+}
+
+onBeforeUnmount(() => {
+  clearCountdown()
+  clearRateLimitCountdown()
+  if (sendDebounceTimer) clearTimeout(sendDebounceTimer)
+})
+</script>
+
+<template>
+  <form @submit.prevent="handleRegister" class="space-y-4">
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">用户名</label>
+      <input
+        v-model="form.username"
+        type="text"
+        required
+        autocomplete="username"
+        placeholder="您的昵称"
+        maxlength="50"
+        class="w-full h-10 px-4 rounded-lg border border-white/[0.08] bg-white/[0.05] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm"
+      />
+    </div>
+
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">邮箱</label>
+      <div class="flex gap-2">
+        <input
+          v-model="form.email"
+          type="email"
+          required
+          autocomplete="email"
+          placeholder="your@email.com"
+          class="flex-1 min-w-0 h-10 px-4 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm"
+        />
+        <button
+          type="button"
+          @click="handleSendCode"
+          :disabled="sendingCode || countdown > 0"
+          class="shrink-0 h-10 px-4 rounded-xl text-xs font-medium transition-all disabled:opacity-50 disabled:cursor-not-allowed"
+          :class="countdown > 0
+            ? 'bg-white/[0.03] text-white/30 border border-white/[0.06]'
+            : 'bg-indigo-500/20 text-indigo-300 border border-indigo-500/20 hover:bg-indigo-500/30'"
+        >
+          <i v-if="sendingCode" class="fas fa-spinner fa-spin"></i>
+          <template v-else-if="countdown > 0">{{ countdown }}s</template>
+          <template v-else>发送验证码</template>
+        </button>
+      </div>
+    </div>
+
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">验证码</label>
+      <input
+        v-model="form.code"
+        type="text"
+        required
+        inputmode="numeric"
+        maxlength="6"
+        placeholder="6 位验证码"
+        autocomplete="one-time-code"
+        class="w-full h-10 px-4 rounded-xl border border-white/[0.08] bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm tracking-widest"
+      />
+    </div>
+
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">密码</label>
+      <div class="relative">
+        <input
+          v-model="form.password"
+          :type="showPwd ? 'text' : 'password'"
+          required
+          autocomplete="new-password"
+          placeholder="至少 6 位"
+          minlength="6"
+          class="w-full h-10 px-4 pr-11 rounded-lg border border-white/[0.08] bg-white/[0.05] text-white placeholder-white/25 focus:outline-none focus:border-indigo-500/50 focus:ring-1 focus:ring-indigo-500/30 transition-all text-sm"
+        />
+        <button
+          type="button"
+          @click="showPwd = !showPwd"
+          class="absolute right-3 top-1/2 -translate-y-1/2 text-white/25 hover:text-white/50 transition-colors"
+        >
+          <i :class="showPwd ? 'fas fa-eye-slash' : 'fas fa-eye'" class="text-xs"></i>
+        </button>
+      </div>
+    </div>
+
+    <div>
+      <label class="block text-sm font-medium text-white/60 mb-2">确认密码</label>
+      <input
+        v-model="form.confirm"
+        type="password"
+        required
+        autocomplete="new-password"
+        placeholder="再次输入密码"
+        class="w-full h-10 px-4 rounded-xl border bg-white/[0.03] text-white placeholder-white/25 focus:outline-none focus:ring-1 transition-all text-sm"
+        :class="passwordMismatch()
+          ? 'border-rose-500/50 focus:border-rose-500/50 focus:ring-rose-500/30'
+          : 'border-white/[0.08] focus:border-indigo-500/50 focus:ring-indigo-500/30'"
+      />
+      <p v-if="passwordMismatch()" class="text-xs text-rose-400 mt-1">两次密码不一致</p>
+    </div>
+
+    <p v-if="error" class="text-sm text-rose-400 flex items-center gap-2">
+      <i class="fas fa-circle-exclamation text-xs"></i>{{ error }}
+    </p>
+    <p v-if="success" class="text-sm text-emerald-400 flex items-center gap-2">
+      <i class="fas fa-circle-check text-xs"></i>{{ success }}
+    </p>
+
+    <BaseButton variant="cta" type="submit" class="w-full" :disabled="loading || passwordMismatch()">
+      <i v-if="loading" class="fas fa-spinner fa-spin text-xs"></i>
+      <span>{{ loading ? '注册中...' : '创建账户' }}</span>
+    </BaseButton>
+  </form>
+</template>
diff --git a/frontend/src/views/home/HomeDemo.vue b/frontend/src/views/home/HomeDemo.vue
new file mode 100644
index 00000000..4c227305
--- /dev/null
+++ b/frontend/src/views/home/HomeDemo.vue
@@ -0,0 +1,71 @@
+<script setup>
+/**
+ * HomeDemo.vue
+ * 落地页产品演示区
+ */
+import { ArrowRight, Sparkles, CheckCircle2, FileDown } from 'lucide-vue-next'
+</script>
+
+<template>
+  <section id="demo" class="relative z-10 py-24 overflow-hidden">
+    <div class="reveal max-w-5xl mx-auto px-4 sm:px-6 relative z-10">
+      <div class="text-center mb-12">
+        <h2 class="reveal text-3xl font-semibold tracking-tight mb-4 text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">眼见为实的转变</h2>
+        <p class="text-white/40 text-sm">复杂的公式、凌乱的涂改，在强大的 AI 引擎面前都不值一提。</p>
+      </div>
+
+      <div class="flex flex-col lg:flex-row gap-4 items-center justify-center">
+        <!-- 左侧：原图模拟 -->
+        <div class="reveal flex-1 w-full max-w-md brand-btn p-6 rounded-lg relative overflow-hidden">
+          <div class="text-xs font-medium tracking-wide text-white/25 uppercase mb-4">Input / 原生图像</div>
+          <div class="bg-[#15151e]/80 p-6 rounded-xl -rotate-1 text-white/50 border border-white/[0.02]">
+            <p class="font-[cursive] text-sm mb-2">3. 巳知函数f(x)=2sin(wx+φ)(w&gt;0,|φ|&lt;π/2)的图像...</p>
+            <p class="font-[cursive] text-sm">A. w=2, φ=π/6</p>
+            <div class="mt-3 w-1/2 h-3 bg-red-400/15 rounded-full blur-[2px] -rotate-3"></div>
+            <p class="font-[cursive] text-xs mt-3 text-red-400/40 font-bold -rotate-6">看不清，选C吧</p>
+          </div>
+        </div>
+
+        <!-- 中间箭头 -->
+        <div class="hidden lg:flex items-center justify-center w-12">
+          <ArrowRight class="w-5 h-5 text-indigo-400/50 animate-pulse" />
+        </div>
+        <div class="lg:hidden flex justify-center py-2">
+          <ArrowRight class="w-5 h-5 text-indigo-400/50 rotate-90 animate-pulse" />
+        </div>
+
+        <!-- 右侧：AI 结果 -->
+        <div class="reveal flex-1 w-full max-w-md brand-btn p-6 rounded-lg relative overflow-hidden text-left" style="transition-delay: 200ms;">
+          <div class="absolute top-0 right-0 w-32 h-32 bg-indigo-500/10 blur-[50px] rounded-full pointer-events-none"></div>
+          
+          <div class="text-xs font-medium tracking-wide text-indigo-400 uppercase flex items-center gap-1.5 mb-4 relative z-10">
+            <Sparkles class="w-3 h-3" /> Output / 结构化数据
+          </div>
+
+          <div class="bg-[#0A0A0F]/80 border border-white/[0.04] p-4 rounded-xl text-white/70 font-mono text-sm leading-relaxed relative z-10 shadow-inner">
+            <div class="flex gap-2 mb-4">
+              <span class="bg-white/[0.04] text-white/50 px-2 py-0.5 rounded-full text-xs border border-white/[0.06]">选择题</span>
+              <span class="bg-white/[0.04] text-white/50 px-2 py-0.5 rounded-full text-xs border border-white/[0.06]">三角函数</span>
+            </div>
+            <p class="text-sm"><span class="text-indigo-400/70 mr-1.5">## 3.</span> 已知函数 <span class="text-indigo-300/80 bg-indigo-500/10 px-1 rounded">$f(x) = 2\sin(\omega x + \varphi)$</span> <span class="text-indigo-300/80 bg-indigo-500/10 px-1 rounded">$(\omega &gt; 0, |\varphi| &lt; \frac{\pi}{2})$</span> 的图像...</p>
+            <div class="mt-4 pl-3 border-l border-white/[0.06] space-y-2">
+              <p class="flex gap-2"><span class="text-white/25">A.</span> <span class="text-white/60">$\omega = 2, \varphi = \frac{\pi}{6}$</span></p>
+              <p class="flex gap-2"><span class="text-white/25">B.</span> <span class="text-white/60">$\omega = 2, \varphi = -\frac{\pi}{6}$</span></p>
+            </div>
+            <div class="mt-6 flex items-center justify-between border-t border-white/[0.06] pt-4">
+              <span class="text-xs text-emerald-400/70 flex items-center gap-1.5">
+                <CheckCircle2 class="w-3 h-3" /> 已过滤手写痕迹
+              </span>
+              <button class="text-xs bg-white/[0.06] text-white/60 hover:bg-white/[0.1] px-3 py-1.5 rounded-lg flex items-center gap-1.5 transition-colors border border-white/[0.06]">
+                导出 MD <FileDown class="w-3 h-3" />
+              </button>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+  </section>
+</template>
diff --git a/frontend/src/views/home/HomeFeatures.vue b/frontend/src/views/home/HomeFeatures.vue
new file mode 100644
index 00000000..53fd885b
--- /dev/null
+++ b/frontend/src/views/home/HomeFeatures.vue
@@ -0,0 +1,133 @@
+<script setup>
+/**
+ * HomeFeatures.vue
+ * 落地页功能特性展示
+ */
+import { onMounted, ref } from 'vue'
+import { Camera, Cpu, Database, BrainCircuit, Sparkles, Layers } from 'lucide-vue-next'
+
+const FEATURES = [
+  {
+    icon: Camera,
+    title: '极速录入，告别手抄',
+    desc: '支持 PDF / 多图并行处理，单文件最高 50 MB，将几十页试卷秒变结构化数据。',
+    span: '', // 普通 1×1
+  },
+  {
+    icon: Cpu,
+    title: '超强公式还原引擎',
+    desc: '基于大模型微调，精准识别手写与印刷体混合的复杂 LaTeX 公式和几何图形。',
+    span: '', // 普通 1×1
+  },
+  {
+    icon: Database,
+    title: 'AI 自动题型分类',
+    desc: '自动判断选择题、填空题、解答题，并剥离答案，生成干净的专属题库。',
+    span: '', // 普通 1×1
+  },
+  {
+    icon: BrainCircuit,
+    title: '错因深度分析',
+    desc: '不是简单的“粗心”，AI 将分析你的每一步计算，指出你的逻辑漏洞与思维盲区。',
+    span: 'md:col-span-2 lg:col-span-1', // 中屏占2列，大屏恢复1列
+  },
+  {
+    icon: Sparkles,
+    title: '同类题智能推演',
+    desc: '掌握一道题，解决一类题。系统会自动生成变式训练，确保知识点真正被消化。',
+    span: 'md:col-span-2', // 中屏及以上占2列
+  },
+  {
+    icon: Layers,
+    title: '多科目知识点打标',
+    desc: '覆盖数学、物理、化学、英语等主流学科，每道题自动关联知识点树，构建个人薄弱点图谱。',
+    span: '',
+  },
+]
+
+// ── 鼠标跟随光晕逻辑 (仅限卡片局部光晕) ──
+
+</script>
+
+<template>
+  <section id="features" class="relative py-24 overflow-hidden">
+
+    <!-- 背景装饰：模糊环境光 (Ambient Glow) -->
+    <div class="absolute top-[20%] left-[-10%] w-[500px] h-[500px] rounded-full bg-indigo-600/5 blur-[150px] pointer-events-none z-0"></div>
+    <div class="absolute bottom-[10%] right-[-5%] w-[400px] h-[400px] rounded-full bg-violet-600/5 blur-[120px] pointer-events-none z-0"></div>
+
+    <!-- 顶部分割线 -->
+    <div class="absolute top-0 left-0 right-0 h-px bg-gradient-to-r from-transparent via-white/[0.06] to-transparent"></div>
+
+    <div class="relative mx-auto max-w-6xl px-4 sm:px-6">
+
+      <!-- 标题区 -->
+      <div class="mb-16 text-center">
+        <h2 class="reveal text-3xl font-semibold tracking-tight mb-4 text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">
+          驱动学习效率的核心引擎
+        </h2>
+        <p class="reveal mx-auto max-w-xl text-sm text-white/35 leading-relaxed">
+          不仅是简单的图像识别，而是真正理解学科内在逻辑的 AI 智能体系统。
+        </p>
+      </div>
+
+      <!-- 特性网格 -->
+      <div class="grid gap-4 md:grid-cols-2 lg:grid-cols-3 relative z-10 max-w-5xl mx-auto">
+        <div
+          v-for="(f, i) in FEATURES"
+          :key="f.title"
+          class="reveal feature-card group relative flex flex-col rounded-[20px] p-[1px] overflow-hidden"
+          :class="f.span"
+          :style="{ transitionDelay: `${i * 80}ms` }"
+        >
+          <!-- 默认状态：灰色 Linear 渐变边框层 -->
+          <div class="absolute inset-0 bg-gradient-to-br from-white/[0.12] via-white/[0.04] to-transparent"></div>
+
+          <!-- 卡片主体背景层 — brand-btn 风格白玻璃 -->
+          <div class="relative h-full w-full rounded-[19px] p-6 flex flex-col items-start text-left transition-all duration-500 overflow-hidden" style="background: rgba(255,255,255,0.05); border: 1px solid rgba(255,255,255,0.08); border-top-color: rgba(255,255,255,0.15); border-bottom-color: rgba(255,255,255,0.03);">
+
+            <!-- 图标容器（BaseLogo 风格） -->
+            <div class="feature-icon relative z-10 mb-4 flex h-10 w-10 items-center justify-center rounded-xl overflow-hidden">
+              <span class="feature-icon__grid absolute inset-0 pointer-events-none"></span>
+              <component :is="f.icon" class="relative h-5 w-5 text-white" />
+            </div>
+            
+            <!-- 文字内容 -->
+            <h3 class="relative z-10 mb-2 text-base font-semibold text-white/90 group-hover:text-white transition-colors duration-300">{{ f.title }}</h3>
+            <p class="relative z-10 text-sm leading-relaxed text-white/40 group-hover:text-white/60 transition-colors duration-300">{{ f.desc }}</p>
+          </div>
+        </div>
+      </div>
+
+    </div>
+  </section>
+</template>
+
+<style scoped>
+/* 移除旧的 hover translateY 浮动效果，保留纯粹的光效反馈 */
+@keyframes gradient-sweep {
+  0% { background-position: -100% center; }
+  100% { background-position: 100% center; }
+}
+
+.animate-gradient-sweep {
+  animation: gradient-sweep 3s ease-in-out infinite alternate;
+}
+
+.feature-icon {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+
+.feature-icon__grid {
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+</style>
diff --git a/frontend/src/views/home/HomeHero.vue b/frontend/src/views/home/HomeHero.vue
new file mode 100644
index 00000000..328016f0
--- /dev/null
+++ b/frontend/src/views/home/HomeHero.vue
@@ -0,0 +1,236 @@
+<script setup>
+/**
+ * HomeHero.vue
+ * 落地页首屏 Hero 区
+ */
+import { ref, onMounted, onUnmounted } from 'vue'
+import { UploadCloud, ArrowRight } from 'lucide-vue-next'
+import BaseButton from '@/components/base/BaseButton.vue'
+import HomePill from '@/components/home/HomePill.vue'
+
+const emit = defineEmits(['scrollToSection'])
+
+// ── 底部星星 ──
+const stars = ref([])
+
+// ── 可见性检测：离开视口时暂停动画 ──
+const heroVisible = ref(true)
+let visibilityObserver = null
+
+onMounted(() => {
+  const count = 40
+  const list = []
+  for (let i = 0; i < count; i++) {
+    list.push({
+      left: Math.random() * 100,
+      top: 60 + Math.random() * 38,
+      size: 1 + Math.random() * 2,
+      opacity: 0.08 + Math.random() * 0.35,
+      delay: Math.random() * 4,
+    })
+  }
+  stars.value = list
+
+  // 监听 Hero 是否在视口内
+  const heroEl = document.getElementById('sticky-hero')
+  if (heroEl) {
+    visibilityObserver = new IntersectionObserver(
+      ([entry]) => { heroVisible.value = entry.isIntersecting },
+      { threshold: 0.05 }
+    )
+    visibilityObserver.observe(heroEl)
+  }
+})
+
+onUnmounted(() => {
+  visibilityObserver?.disconnect()
+})
+</script>
+
+<template>
+  <!-- ① Sticky Hero 容器 — Linear 风格 -->
+  <div id="sticky-hero" class="sticky top-0 h-screen overflow-hidden z-0 flex flex-col justify-center bg-[#0A0A0F]" style="contain: content;">
+
+    <!-- 背景装饰：复杂流体拓扑波纹 (Fluid Topography) 基础层 -->
+    <div class="absolute inset-0 pointer-events-none z-0 opacity-20" style="
+      background-image: url('data:image/svg+xml,%3Csvg viewBox=%220 0 1000 1000%22 xmlns=%22http://www.w3.org/2000/svg%22%3E%3Cfilter id=%22noiseFilter%22%3E%3CfeTurbulence type=%22fractalNoise%22 baseFrequency=%220.005%22 numOctaves=%223%22 stitchTiles=%22stitch%22/%3E%3CfeColorMatrix type=%22matrix%22 values=%221 0 0 0 0, 0 1 0 0 0, 0 0 1 0 0, 0 0 0 10 -4%22 /%3E%3C/filter%3E%3Crect width=%22100%25%22 height=%22100%25%22 filter=%22url(%23noiseFilter)%22 fill=%22none%22 stroke=%22%23ffffff%22 stroke-width=%221%22 opacity=%220.3%22/%3E%3Cpath d=%22M0,100 C200,300 300,0 500,100 C700,200 800,-100 1000,100 M0,200 C250,400 350,100 550,200 C750,300 850,0 1000,200 M0,300 C300,500 400,200 600,300 C800,400 900,100 1000,300 M0,400 C350,600 450,300 650,400 C850,500 950,200 1000,400 M0,500 C400,700 500,400 700,500 C900,600 1000,300 1000,500 M0,600 C450,800 550,500 750,600 C950,700 1000,400 1000,600 M0,700 C500,900 600,600 800,700 C1000,800 1000,500 1000,700 M0,800 C550,1000 650,700 850,800 C1000,900 1000,600 1000,800 M0,900 C600,1100 700,800 900,900 C1000,1000 1000,700 1000,900%22 stroke=%22%23ffffff%22 stroke-width=%221%22 fill=%22none%22 opacity=%220.15%22 /%3E%3C/svg%3E');
+      background-size: cover;
+      background-position: center;
+      mask-image: radial-gradient(ellipse 100% 100% at 50% 50%, black, transparent);
+      -webkit-mask-image: radial-gradient(ellipse 100% 100% at 50% 50%, black, transparent);
+    "></div>
+
+    <!-- 顶部装饰椭圆与土星光环 — 用 vw 统一大小和位置的缩放基准 -->
+    <div class="absolute pointer-events-none z-0 flex items-center justify-center" style="
+      top: -100vw;
+      left: 50%;
+      transform: translateX(-50%);
+      width: 140vw;
+      aspect-ratio: 1.3 / 1;
+    ">
+      <!-- 椭圆本体 -->
+      <div class="absolute inset-0" style="
+        border-radius: 50%;
+        background: linear-gradient(to bottom, rgba(92,81,148,0.5), rgba(47,40,91,0.95));
+        box-shadow:
+          inset 0 -20px 24px 0 rgba(255,255,255,0.15),
+          0 16px 32px 0 rgba(97,62,210,0.32),
+          inset 0 -1px 0 0 rgba(129,115,223,0.6);
+      "></div>
+
+      <!-- 土星光环: LaTeX 公式 -->
+      <!-- 放大容器以远离"土星"本体，同时增加 Z 轴旋转(rotateX)和倾斜(rotateZ)产生悬浮交错的纵深感 -->
+      <div class="absolute pointer-events-none" :class="{ 'rings-paused': !heroVisible }" style="
+        width: 120%;
+        height: 120%;
+        top: -10%;
+        left: -10%;
+        transform-origin: center center;
+        transform: rotateX(15deg) rotateZ(-4deg) translateY(5%);
+      ">
+        <svg viewBox="-100 -100 1200 1200" preserveAspectRatio="none" class="w-full h-full opacity-60" style="overflow: visible;">
+          <!-- 第一圈光环 (内圈，顺时针旋转) -->
+          <g class="ring-spin-cw" style="transform-origin: 500px 500px;">
+            <!-- 光环轨道 (反向绘制路径以使文字正向朝外) -->
+            <path id="saturn-ring-path-1" d="M 500, 980 A 480,480 0 1,0 500,20 A 480,480 0 1,0 500,980" fill="none" />
+            
+            <!-- 轨道装饰线 -->
+            <path d="M 500, 500 m -488, 0 a 488,488 0 1,1 976,0 a 488,488 0 1,1 -976,0" fill="none" stroke="rgba(129,115,223,0.2)" stroke-width="1" />
+            <path d="M 500, 500 m -472, 0 a 472,472 0 1,1 944,0 a 472,472 0 1,1 -944,0" fill="none" stroke="rgba(129,115,223,0.2)" stroke-width="1" />
+
+            <!-- 公式文本 -->
+            <text fill="currentColor" class="text-indigo-200/80" font-family="'Times New Roman', Times, serif" font-size="18" font-style="italic" letter-spacing="5">
+              <textPath href="#saturn-ring-path-1" startOffset="0%">
+                ∇⋅E = ρ/ε₀ &nbsp;&nbsp;&nbsp;&nbsp; ∇×B = μ₀J + μ₀ε₀(∂E/∂t) &nbsp;&nbsp;&nbsp;&nbsp; R_μν - ½Rg_μν + Λg_μν = (8πG/c⁴)T_μν &nbsp;&nbsp;&nbsp;&nbsp; iℏ(∂Ψ/∂t) = ĤΨ &nbsp;&nbsp;&nbsp;&nbsp; (iγ^μ∂_μ - m)ψ = 0 &nbsp;&nbsp;&nbsp;&nbsp; e^(iπ) + 1 = 0 &nbsp;&nbsp;&nbsp;&nbsp; S = ∫ L dt &nbsp;&nbsp;&nbsp;&nbsp; ∇⋅E = ρ/ε₀ &nbsp;&nbsp;&nbsp;&nbsp; ∇×B = μ₀J + μ₀ε₀(∂E/∂t) &nbsp;&nbsp;&nbsp;&nbsp; R_μν - ½Rg_μν + Λg_μν = (8πG/c⁴)T_μν &nbsp;&nbsp;&nbsp;&nbsp; iℏ(∂Ψ/∂t) = ĤΨ &nbsp;&nbsp;&nbsp;&nbsp; (iγ^μ∂_μ - m)ψ = 0 &nbsp;&nbsp;&nbsp;&nbsp; e^(iπ) + 1 = 0 &nbsp;&nbsp;&nbsp;&nbsp; S = ∫ L dt &nbsp;&nbsp;&nbsp;&nbsp; ∇⋅E = ρ/ε₀ &nbsp;&nbsp;&nbsp;&nbsp; ∇×B = μ₀J + μ₀ε₀(∂E/∂t) &nbsp;&nbsp;&nbsp;&nbsp; R_μν - ½Rg_μν + Λg_μν = (8πG/c⁴)T_μν &nbsp;&nbsp;&nbsp;&nbsp; iℏ(∂Ψ/∂t) = ĤΨ &nbsp;&nbsp;&nbsp;&nbsp; (iγ^μ∂_μ - m)ψ = 0 &nbsp;&nbsp;&nbsp;&nbsp; e^(iπ) + 1 = 0 &nbsp;&nbsp;&nbsp;&nbsp; S = ∫ L dt
+              </textPath>
+            </text>
+          </g>
+
+          <!-- 第二圈光环 (外圈，逆时针旋转，半径更大，文本不同) -->
+          <g class="ring-spin-ccw" style="transform-origin: 500px 500px;">
+            <!-- 光环轨道 (半径从 480 增加到 540) -->
+            <path id="saturn-ring-path-2" d="M 500, 1040 A 540,540 0 1,0 500,-40 A 540,540 0 1,0 500,1040" fill="none" />
+            
+            <!-- 轨道装饰线 -->
+            <path d="M 500, 500 m -546, 0 a 546,546 0 1,1 1092,0 a 546,546 0 1,1 -1092,0" fill="none" stroke="rgba(129,115,223,0.15)" stroke-width="1" />
+            <path d="M 500, 500 m -534, 0 a 534,534 0 1,1 1068,0 a 534,534 0 1,1 -1068,0" fill="none" stroke="rgba(129,115,223,0.15)" stroke-width="1" />
+
+            <!-- 公式文本 (换成另一组数学公式：傅里叶变换、高斯-博内定理、薛定谔、贝叶斯等) -->
+            <text fill="currentColor" class="text-indigo-200/70" font-family="'Times New Roman', Times, serif" font-size="14" font-style="italic" letter-spacing="6">
+              <textPath href="#saturn-ring-path-2" startOffset="0%">
+                F(ω) = ∫ f(t)e^(-iωt)dt &nbsp;&nbsp;&nbsp;&nbsp; ∫K dA + ∫k_g ds = 2πχ(M) &nbsp;&nbsp;&nbsp;&nbsp; P(A|B) = P(B|A)P(A)/P(B) &nbsp;&nbsp;&nbsp;&nbsp; dS ≥ 0 &nbsp;&nbsp;&nbsp;&nbsp; E = mc² &nbsp;&nbsp;&nbsp;&nbsp; V - E + F = 2 &nbsp;&nbsp;&nbsp;&nbsp; F(ω) = ∫ f(t)e^(-iωt)dt &nbsp;&nbsp;&nbsp;&nbsp; ∫K dA + ∫k_g ds = 2πχ(M) &nbsp;&nbsp;&nbsp;&nbsp; P(A|B) = P(B|A)P(A)/P(B) &nbsp;&nbsp;&nbsp;&nbsp; dS ≥ 0 &nbsp;&nbsp;&nbsp;&nbsp; E = mc² &nbsp;&nbsp;&nbsp;&nbsp; V - E + F = 2 &nbsp;&nbsp;&nbsp;&nbsp; F(ω) = ∫ f(t)e^(-iωt)dt &nbsp;&nbsp;&nbsp;&nbsp; ∫K dA + ∫k_g ds = 2πχ(M) &nbsp;&nbsp;&nbsp;&nbsp; P(A|B) = P(B|A)P(A)/P(B) &nbsp;&nbsp;&nbsp;&nbsp; dS ≥ 0 &nbsp;&nbsp;&nbsp;&nbsp; E = mc² &nbsp;&nbsp;&nbsp;&nbsp; V - E + F = 2 &nbsp;&nbsp;&nbsp;&nbsp; F(ω) = ∫ f(t)e^(-iωt)dt &nbsp;&nbsp;&nbsp;&nbsp; ∫K dA + ∫k_g ds = 2πχ(M) &nbsp;&nbsp;&nbsp;&nbsp; P(A|B) = P(B|A)P(A)/P(B) &nbsp;&nbsp;&nbsp;&nbsp; dS ≥ 0 &nbsp;&nbsp;&nbsp;&nbsp; E = mc² &nbsp;&nbsp;&nbsp;&nbsp; V - E + F = 2
+              </textPath>
+            </text>
+          </g>
+        </svg>
+      </div>
+    </div>
+
+    <!-- 装饰: 光斑 -->
+    <div class="absolute pointer-events-none z-0 w-80 h-80 rounded-full blur-[120px] bg-indigo-600/[0.07]"
+      style="top: 25%; left: 8%;"
+    ></div>
+    <div class="absolute pointer-events-none z-0 w-64 h-64 rounded-full blur-[100px] bg-violet-500/[0.05]"
+      style="top: 55%; right: 10%;"
+    ></div>
+
+    <!-- 装饰: 底部简约的环境光晕 -->
+    <div class="absolute bottom-0 left-1/2 -translate-x-1/2 w-[80vw] h-[30vh] pointer-events-none z-0">
+      <div class="absolute inset-0 bg-indigo-500/10 blur-[100px] rounded-[100%] transform translate-y-1/2"></div>
+      <div class="absolute inset-0 bg-violet-500/5 blur-[120px] rounded-[100%] transform translate-y-1/2 scale-150"></div>
+    </div>
+
+    <!-- 首屏区块 — 居中布局 -->
+    <section id="hero" class="relative w-full px-4 sm:px-6 lg:px-8 max-w-6xl mx-auto z-10">
+      <div class="text-center">
+        <!-- 标签 -->
+        <HomePill class="mb-6" />
+
+        <h1 class="text-4xl sm:text-5xl font-semibold tracking-tight leading-tight mb-6 text-white">
+          重塑错题整理<br />
+          <span class="text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">一键生成知识图谱</span>
+        </h1>
+
+        <p class="text-base text-white/40 mb-8 max-w-lg mx-auto leading-relaxed">
+          上传试卷或手写笔记，AI 自动完成 OCR 识别、题目分割、公式还原、知识点标注。
+        </p>
+
+        <div class="flex gap-3 justify-center">
+          <BaseButton to="/auth">
+            <UploadCloud class="w-4 h-4" />
+            开始使用
+          </BaseButton>
+          <BaseButton variant="secondary" href="#demo">
+            查看演示
+            <ArrowRight class="w-4 h-4" />
+          </BaseButton>
+        </div>
+      </div>
+    </section>
+
+    <!-- 星星点缀 -->
+    <div class="absolute inset-0 pointer-events-none z-[1]">
+      <div
+        v-for="(s, i) in stars"
+        :key="i"
+        class="absolute rounded-full bg-white"
+        :style="{
+          left: s.left + '%',
+          top: s.top + '%',
+          width: s.size + 'px',
+          height: s.size + 'px',
+          opacity: s.opacity,
+        }"
+      ></div>
+    </div>
+
+    <!-- 底部箭头 -->
+    <button
+      @click="emit('scrollToSection', 'features')"
+      class="absolute bottom-8 left-1/2 -translate-x-1/2 z-20 text-white/20 hover:text-white/50 transition-colors cursor-pointer"
+    >
+      <i class="fa-solid fa-chevron-down text-lg animate-bounce"></i>
+    </button>
+  </div>
+</template>
+
+<style scoped>
+@keyframes spin-cw {
+  0% { transform: rotate(0deg) translateZ(0); }
+  100% { transform: rotate(360deg) translateZ(0); }
+}
+
+@keyframes spin-ccw {
+  0% { transform: rotate(360deg) translateZ(0); }
+  100% { transform: rotate(0deg) translateZ(0); }
+}
+
+.ring-spin-cw {
+  animation: spin-cw 120s linear infinite;
+  will-change: transform;
+}
+
+.ring-spin-ccw {
+  animation: spin-ccw 180s linear infinite;
+  will-change: transform;
+}
+
+/* 离开视口时暂停旋转，节省渲染开销 */
+.rings-paused .ring-spin-cw,
+.rings-paused .ring-spin-ccw {
+  animation-play-state: paused;
+}
+
+@keyframes gradient-sweep {
+  0% { background-position: -100% center; }
+  100% { background-position: 100% center; }
+}
+
+.animate-gradient-sweep {
+  animation: gradient-sweep 3s ease-in-out infinite alternate;
+}
+
+</style>
diff --git a/frontend/src/views/home/HomeWorkflow.vue b/frontend/src/views/home/HomeWorkflow.vue
new file mode 100644
index 00000000..373a1edc
--- /dev/null
+++ b/frontend/src/views/home/HomeWorkflow.vue
@@ -0,0 +1,126 @@
+<script setup>
+/**
+ * HomeWorkflow.vue
+ * 落地页工作流程展示
+ */
+import { ref, onMounted, onUnmounted } from 'vue'
+import { UploadCloud, BrainCircuit, Sparkles, FileDown } from 'lucide-vue-next'
+
+const STEPS = [
+  { icon: 'upload-cloud', num: '01', title: '上传试卷',   desc: '支持PDF/多图并行处理',  delay: 0 },
+  { icon: 'brain-circuit', num: '02', title: 'AI 解析拆题', desc: '智能排版与图文切分',    delay: 150 },
+  { icon: 'sparkles',     num: '03', title: '纠错与打标', desc: '公式还原与知识点关联',   delay: 300 },
+  { icon: 'file-down',    num: '04', title: '一键导出',   desc: '生成 Markdown/PDF',      delay: 450 },
+]
+
+const iconMap = {
+  'upload-cloud': UploadCloud,
+  'brain-circuit': BrainCircuit,
+  sparkles: Sparkles,
+  'file-down': FileDown,
+}
+
+const activeStep = ref(0)
+let intervalId = null
+
+function startStepInterval() {
+  intervalId = setInterval(() => {
+    activeStep.value = (activeStep.value + 1) % STEPS.length
+  }, 3000)
+}
+
+function onStepClick(idx) {
+  clearInterval(intervalId)
+  activeStep.value = idx
+  startStepInterval()
+}
+
+function getStepIconClass(idx) {
+  if (idx <= activeStep.value) {
+    return 'wf-icon wf-icon--active w-12 h-12 rounded-xl flex items-center justify-center mb-4 text-white transition-all duration-500'
+  }
+  return 'w-12 h-12 rounded-xl flex items-center justify-center mb-4 bg-white/[0.04] text-white/25 border border-white/[0.06] transition-all duration-500'
+}
+
+function getStepContainerClass(idx) {
+  if (idx === activeStep.value) {
+    return 'relative p-4 rounded-lg cursor-pointer brand-btn transition-all duration-500 transform scale-105'
+  }
+  return 'relative p-4 rounded-lg cursor-pointer border border-transparent hover:bg-white/[0.03] hover:border-white/[0.05] transition-all duration-500'
+}
+
+function getStepTitleClass(idx) {
+  if (idx === activeStep.value) return 'text-sm font-semibold mb-1 text-white/90'
+  return 'text-sm font-semibold mb-1 text-white/40'
+}
+
+function getStepDescClass(idx) {
+  if (idx === activeStep.value) return 'text-xs text-white/50'
+  return 'text-xs text-white/25'
+}
+
+function getProgressWidth() {
+  return `${(activeStep.value / (STEPS.length - 1)) * 100}%`
+}
+
+onMounted(() => {
+  startStepInterval()
+})
+
+onUnmounted(() => {
+  if (intervalId) clearInterval(intervalId)
+})
+</script>
+
+<template>
+  <section id="workflow" class="relative z-10 py-24 overflow-hidden">
+    <div class="reveal max-w-5xl mx-auto px-4 sm:px-6 relative z-10">
+      <div class="text-center mb-16">
+        <h2 class="reveal text-3xl font-semibold tracking-tight mb-4 text-transparent bg-clip-text animate-gradient-sweep" style="
+            background-image: linear-gradient(to right, rgb(151, 137, 222) 0%, rgb(151, 137, 222) 20%, rgb(255, 255, 255) 50%, rgb(151, 137, 222) 80%, rgb(151, 137, 222) 100%);
+            background-size: 200% auto;
+          ">极简四步，自动运转</h2>
+        <p class="text-white/40 text-sm">将原本需要耗费数小时的繁杂抄录，浓缩进点击之间。</p>
+      </div>
+
+      <div class="relative max-w-4xl mx-auto">
+        <div class="grid grid-cols-1 lg:grid-cols-4 gap-4 relative z-10">
+          <div
+            v-for="(s, i) in STEPS"
+            :key="s.num"
+            :class="getStepContainerClass(i)"
+            :data-index="i"
+            @click="onStepClick(i)"
+          >
+            <div class="flex flex-col items-center text-center">
+              <div :class="getStepIconClass(i)" class="relative overflow-hidden">
+                <span v-if="i <= activeStep" class="wf-icon__grid absolute inset-0 pointer-events-none"></span>
+                <component :is="iconMap[s.icon]" class="relative w-5 h-5" />
+              </div>
+              <h4 :class="getStepTitleClass(i)">
+                {{ s.title }}
+              </h4>
+              <p :class="getStepDescClass(i)">{{ s.desc }}</p>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+  </section>
+</template>
+
+<style scoped>
+.wf-icon--active {
+  background: linear-gradient(to bottom, rgba(129, 115, 223, 0.9), rgba(99, 87, 199, 0.9));
+  box-shadow: inset 0 1px 0 0 rgba(255, 255, 255, 0.12);
+}
+
+.wf-icon__grid {
+  background-image:
+    linear-gradient(to right, rgba(255, 255, 255, 0.06) 1px, transparent 1px),
+    linear-gradient(to bottom, rgba(255, 255, 255, 0.06) 1px, transparent 1px);
+  background-size: 8px 8px;
+  mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+  -webkit-mask-image: radial-gradient(ellipse at center, black 30%, transparent 80%);
+}
+</style>
diff --git a/frontend/tailwind.config.js b/frontend/tailwind.config.js
index 2b0bec19..b9217667 100644
--- a/frontend/tailwind.config.js
+++ b/frontend/tailwind.config.js
@@ -1,4 +1,6 @@
 /** @type {import('tailwindcss').Config} */
+import typography from '@tailwindcss/typography'
+
 export default {
   // 在 content 数组中补充 "./app.html"
   content: ["./index.html", "./app.html", "./src/**/*.{vue,js,ts,jsx,tsx}"],
@@ -6,5 +8,7 @@ export default {
   theme: {
     extend: {},
   },
-  plugins: [],
+  plugins: [
+    typography,
+  ],
 }
\ No newline at end of file
diff --git a/frontend/vite.config.js b/frontend/vite.config.js
index d09f2a32..17234f8e 100644
--- a/frontend/vite.config.js
+++ b/frontend/vite.config.js
@@ -3,7 +3,12 @@ import vue from '@vitejs/plugin-vue'
 import { resolve } from 'path'
 
 // https://vite.dev/config/
-export default defineConfig(({ command }) => ({
+export default defineConfig({
+  resolve: {
+    alias: {
+      '@': resolve(__dirname, 'src'),
+    },
+  },
   plugins: [
     vue(),
     // 开发模式：将 /auth 和 /app 路径重写到 app.html（SPA 入口）
@@ -12,7 +17,7 @@ export default defineConfig(({ command }) => ({
       configureServer(server) {
         server.middlewares.use((req, res, next) => {
           const url = req.url?.split('?')[0]
-          if (url === '/' || url === '/auth' || url === '/app' || url?.startsWith('/app/')) {
+          if (url === '/' || url === '/auth' || url === '/auth/login' || url === '/auth/register' || url === '/app' || url?.startsWith('/app/')) {
             req.url = '/app.html'
           }
           next()
@@ -24,6 +29,7 @@ export default defineConfig(({ command }) => ({
     environment: 'jsdom',
   },
   server: {
+    host: '127.0.0.1',
     proxy: {
       '/api': {
         target: 'http://localhost:5001',
@@ -37,10 +43,17 @@ export default defineConfig(({ command }) => ({
         target: 'http://localhost:5001',
         changeOrigin: true,
       },
+      '/erased': {
+        target: 'http://localhost:5001',
+        changeOrigin: true,
+      },
+      '/uploads': {
+        target: 'http://localhost:5001',
+        changeOrigin: true,
+      },
     },
   },
-  // 开发模式用 /，构建时才加 /static/vue/ 前缀给 Flask 托管
-  base: command === 'build' ? '/static/vue/' : '/',
+  base: '/',
   build: {
     outDir: 'dist',
     emptyOutDir: true,
@@ -52,4 +65,4 @@ export default defineConfig(({ command }) => ({
       }
     }
   },
-}))
\ No newline at end of file
+})
\ No newline at end of file
diff --git a/requirements.txt b/requirements.txt
index b9c90fed..7a5cfbfe 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -42,6 +42,7 @@ sqlalchemy>=2.0
 
 # --- Web 框架 ---
 flask==3.1.2
+flask-cors==6.0.2
 werkzeug==3.1.5
 aiohttp>=3.9.0
 
diff --git "a/\345\211\215\347\253\257\345\274\200\345\217\221\350\247\204\350\214\203.md" "b/\345\211\215\347\253\257\345\274\200\345\217\221\350\247\204\350\214\203.md"
deleted file mode 100644
index 6f5bb10e..00000000
--- "a/\345\211\215\347\253\257\345\274\200\345\217\221\350\247\204\350\214\203.md"
+++ /dev/null
@@ -1,289 +0,0 @@
-# 前端开发规范
-
-> 本文档从 `CLAUDE.md` 中提取并整理，作为前端开发的完整参考手册。
-
----
-
-## 1. 技术栈
-
-| 类别 | 选型 | 备注 |
-|------|------|------|
-| 框架 | Vue 3 + `<script setup>` | Composition API，禁止 Options API |
-| 构建 | Vite | 多页面入口：`index.html`（介绍页）+ `app.html`（工作台） |
-| 样式 | Tailwind CSS 3 | utility-first，class-based dark mode |
-| 交互组件 | HeadlessUI Vue | 用于可访问的弹窗、下拉等 |
-| HTML 净化 | DOMPurify | 白名单在 `utils.js` 的 `ALLOWED_HTML_TAGS` |
-| 语言 | 纯 JavaScript | 不使用 TypeScript |
-| 网络请求 | 原生 fetch | 不使用 axios，文件上传用 XHR |
-| 状态管理 | 无 | 全局状态集中在 `App.vue`（ref/reactive/computed/watch） |
-| CDN 依赖 | Font Awesome 6.5、MathJax 3、SortableJS 1.15、Chart.js 4 | 非 npm 安装 |
-
----
-
-## 2. 布局规范
-
-### PC 端
-- 左侧固定侧边栏：`aside w-64`
-- 右侧主内容区：flex-1
-
-### 移动端
-- 底部 Tab 导航栏：`fixed bottom-0`
-- 全宽内容区
-
-### 视图切换
-- `currentView` ref 控制视图（`v-show` 切换，非 `v-if`）
-- 路由映射在 `VIEW_TO_PATH` 对象中集中管理
-
----
-
-## 3. 间距节奏（8pt Grid）
-
-**所有 padding / margin / gap 只允许使用 4 的倍数**，对应 Tailwind 档位：
-
-| 用途 | 允许值（Tailwind） | 像素 |
-|------|-------------------|------|
-| 微间距（图标、标签内边距） | `p-1` `p-2` `gap-1` `gap-2` | 4px、8px |
-| 小间距（按钮、行内元素） | `p-3` `p-4` `gap-3` `gap-4` | 12px、16px |
-| 中间距（卡片内边距、区块间距） | `p-6` `p-8` `gap-6` `gap-8` | 24px、32px |
-| 大间距（页面边距、区域分割） | `p-10` `p-12` `gap-10` `gap-12` | 40px、48px |
-
-### 禁止使用
-
-```
-❌ p-5  p-7  p-9  gap-5  gap-7  gap-9
-❌ px-[17px]  py-[13px]  gap-[22px]  （任意魔法数字）
-❌ space-y-3  space-y-5  space-y-7  （非4倍数的 space 值）
-```
-
-### 正确示例
-
-```html
-<!-- ✅ 正确 -->
-<div class="p-6 gap-4">
-<div class="mb-8 px-4 py-2">
-
-<!-- ❌ 错误 -->
-<div class="p-5 gap-3">
-<div class="mb-7 px-[15px]">
-```
-
----
-
-## 4. 排版层级（Typography）
-
-字号只允许以下 **7 档**，禁止使用其他任何值：
-
-| 层级 | Tailwind | 像素 | 行高 | 用途 |
-|------|----------|------|------|------|
-| 超大标题 | `text-4xl` | 36px | `leading-tight`（1.25） | 落地页 Hero、各区块主标题 |
-| 大标题 | `text-3xl` | 30px | `leading-tight`（1.25） | 页面主标题、工作台区块标题 |
-| 标题 | `text-2xl` | 24px | `leading-snug`（1.375） | 区块标题、Modal 标题 |
-| 副标题 | `text-xl` | 20px | `leading-snug` | 卡片标题、侧边栏项 |
-| 正文 | `text-base` | 16px | `leading-relaxed`（1.625） | 题目内容、说明文字 |
-| 辅助 | `text-sm` | 14px | `leading-relaxed` | 标签、次要信息 |
-| 标注 | `text-xs` | 12px | `leading-normal`（1.5） | 元数据、时间戳、徽标 |
-
-### 禁止使用
-
-```
-❌ text-lg  text-[10px]  text-[11px]  text-[13px]  text-[15px]
-❌ 任何 text-[Npx] 自定义值
-```
-
-### 信息密度原则
-
-每张 Card 只允许 **1 个主信息 + 2 个辅信息**，禁止堆砌超过 3 层文字层级。
-
----
-
-## 5. 配色 Token
-
-**只允许以下色系，禁止随意引入彩虹色。**
-
-### 语义色
-
-| 语义 | 亮色 | 暗色 |
-|------|------|------|
-| 页面背景 | `white` / `slate-50` | `slate-900` / `slate-950` / `[#0A0A0F]` |
-| 卡片面（玻璃态） | `bg-white/70 backdrop-blur-xl` | `bg-white/[0.03] backdrop-blur-xl` |
-| 边框 | `border-slate-200/60` | `border-white/10` |
-| 主操作 | `blue-600` / `indigo-600` | `indigo-500` |
-| 成功 | `emerald-600` | `emerald-400` |
-| 错误 | `rose-600` | `rose-400` |
-| 中性文字 | `slate-900` / `slate-700` / `slate-500` / `slate-400` | `white` / `slate-200` / `slate-400` / `slate-500` |
-
-### 禁止使用的颜色
-
-```
-❌ orange  amber  fuchsia  pink  cyan  teal  lime  yellow  violet  purple
-```
-
-> 如果需要状态区分（如"复习中"），使用 `indigo`（主色）或 `slate`（中性色），不要用 amber/orange。
-
-### 玻璃态卡片标准写法
-
-```html
-<!-- 亮色 -->
-<div class="rounded-2xl border border-slate-200/60 bg-white/70 p-6 shadow-sm backdrop-blur-xl">
-
-<!-- 暗色 -->
-<div class="dark:border-white/10 dark:bg-white/[0.03]">
-```
-
----
-
-## 6. 圆角 Token
-
-| 元素 | 值 | 示例 |
-|------|-----|------|
-| 大卡片、Modal、弹窗 | `rounded-2xl` / `rounded-3xl` | 内容容器、对话框 |
-| 按钮、输入框、小卡片 | `rounded-xl` | 操作按钮、表单控件 |
-| 标签、Badge、Pill | `rounded-full` | 学科标签、状态徽标、数量指示器 |
-| 图标容器、小元素 | `rounded-lg` | 复选框、头像容器 |
-
-### 禁止使用
-
-```
-❌ rounded-[2rem]  rounded-md  rounded-sm  rounded-none
-❌ 任何 rounded-[Npx] 自定义值
-```
-
----
-
-## 7. 阴影层级
-
-**只允许 2 层，禁止滥用大投影。**
-
-| 状态 | 值 | 用途 |
-|------|-----|------|
-| 默认 | `shadow-sm` | 卡片、输入框的静息态 |
-| Hover / 激活 | `shadow-md` 或品牌色光晕 `shadow-blue-500/20` | 交互反馈 |
-
-### 禁止使用
-
-```
-❌ shadow-lg  shadow-xl  shadow-2xl  shadow-inner
-```
-
----
-
-## 8. 按钮样式
-
-### 尺寸
-
-- 高度统一 `h-10`（40px）
-- 图标尺寸 `size-4`（16px）或 `size-5`（20px）
-
-### 类型
-
-| 类型 | 样式 |
-|------|------|
-| 主按钮 | `h-10 rounded-xl bg-blue-600 text-white text-sm font-bold` + hover 光晕 |
-| 成功按钮 | `h-10 rounded-xl border border-emerald-500/30 bg-emerald-50/80 text-emerald-700 text-sm font-bold` |
-| 次按钮 | `h-10 rounded-xl border border-slate-200/60 bg-white/60 text-slate-700 text-sm font-bold` |
-
-### 通用属性
-
-```html
-<!-- 每个按钮都必须包含 -->
-class="inline-flex items-center justify-center gap-2 transition-all"
-
-<!-- 禁用态 -->
-class="disabled:cursor-not-allowed disabled:opacity-50"
-```
-
-### 禁止
-
-```
-❌ h-11  h-12  h-9  h-8  （非标准高度）
-❌ hover:-translate-y-0.5  （按钮不应上移）
-```
-
----
-
-## 9. 暗色模式
-
-- **所有元素必须包含 `dark:` 变体**，不能只写亮色样式
-- 暗色背景使用 `slate-900` / `slate-950` / `[#0A0A0F]`
-- 主题切换支持 View Transitions API 圆形扩散动画
-- `localStorage.theme` 持久化用户偏好
-
-### 暗色模式检查清单
-
-```
-✅ 背景色有 dark: 变体
-✅ 文字颜色有 dark: 变体
-✅ 边框颜色有 dark: 变体
-✅ hover 状态有 dark: 变体
-✅ focus 状态有 dark: 变体
-```
-
----
-
-## 10. 组件规范
-
-### 文件组织
-
-- 新功能提取为独立 `.vue` 组件放 `src/components/`
-- 关键 DOM 元素添加语义类名或 `data-testid`
-- 父子通信：props 向下，emits 向上
-- UI 文案使用中文
-- API 路径前缀 `/api/`
-
-### 已抽离的公共组件
-
-| 组件 | 用途 |
-|------|------|
-| `GlassCard` | 玻璃态卡片容器，支持 `padding` `rounded` props |
-| `PageHeader` | 页面标题区，支持徽章、标题、副标题、`#extra` 插槽 |
-| `StatCard` | 统计数字卡片，支持 `label` `icon` `value` `unit` `color` |
-| `QuestionItem` | 题目列表项，支持选择、状态显示、`#extra` `#actions` 插槽 |
-| `ProviderSection` | API Provider 配置区块（标题 + 列表项 + 空状态） |
-| `SearchInput` | 搜索输入框，支持 `v-model` `label` `placeholder` `icon` |
-| `CustomSelect` | 下拉选择框（玻璃态） |
-| `CalendarPicker` | 日期选择器 |
-| `SplitLoading` | AI 分割任务全屏遮罩动画 |
-
-### 组件 props 命名
-
-- 布尔值用 `is` / `show` / `has` 前缀：`showStatus` `selectable`
-- 事件用动词：`@click` `@toggle-select` `@remove`
-- 颜色类 prop 使用 Tailwind 颜色名：`color="indigo"`
-
----
-
-## 11. 交互规范
-
-### Hover 效果
-
-- 卡片：`hover:shadow-md`（阴影加深，不位移）
-- 按钮：颜色/透明度变化 + `transition-all`
-- **禁止** `hover:-translate-y` 上移效果（除非是落地页动画）
-
-### 过渡动画
-
-- 所有可交互元素必须包含 `transition-all` 或具体的 transition 属性
-- 视图切换使用 `<Transition name="view-fade">`
-- 弹窗使用 HeadlessUI 内置过渡
-
-### 下拉面板
-
-- 面板背景不透明：亮色 `bg-white`，暗色 `bg-slate-900`
-- 避免透明穿透导致文字重叠
-
----
-
-## 12. 快速自检清单
-
-开发新组件或修改现有组件时，逐项检查：
-
-- [ ] 间距是否为 4 的倍数（无 p-5、gap-5 等）
-- [ ] 字号是否在 7 档之内（无 text-[10px] 等）
-- [ ] 配色是否在允许色系内（无 orange、amber 等）
-- [ ] 圆角是否符合 Token（标签用 rounded-full，卡片用 rounded-2xl）
-- [ ] 阴影是否只用 shadow-sm / shadow-md
-- [ ] 按钮高度是否为 h-10
-- [ ] 所有样式是否有 dark: 变体
-- [ ] 是否有不必要的 hover 位移
-- [ ] 是否有 transition-all 或过渡动画
-- [ ] Card 信息层级是否不超过 3 层