A Claude Code skill for deep, data-rich World Cup match analysis — squads, injuries, coaches, referee tendencies, venue / altitude / weather, plus calibrated win-draw-loss and scoreline probabilities.
Entertainment only — not betting advice.
🌐 Live website: wc.netzfs.com · source: netz888/WORLD-CUP-AI-HUB (Next.js). 在线网站：wc.netzfs.com · 源码： netz888/WORLD-CUP-AI-HUB（Next.js）。
Sub-project 1 (self-contained skill) is implemented. The skill, its references, templates,
the analysis-JSON schema, a trained Elo multi-factor prediction engine (tools/predict.mjs),
a zero-dependency validator, and a full Brazil-vs-Morocco sample are all in the repo. See the
design spec in docs/superpowers/specs/.
Probabilities come from a v6 Elo multi-factor engine (worldcup-deep-analysis/tools/predict.mjs),
grid-searched + leave-one-tournament-out cross-validated on 192 real World Cup group-stage matches
(2010 / 2014 / 2018 / 2022). Train vs validation exact-score hit rate is ~16.1% vs ~16.7%
(no overfitting); 1X2 direction ~55%. Exact-score accuracy has a physical ceiling (~16–17%) — the
engine is honest about it. Details: references/prediction-model.md.
The skill is self-contained — no paid API, no keys. Node runs the trained prediction engine and the validator.
- Copy the skill into your Claude Code skills directory:
cp -r worldcup-deep-analysis ~/.claude/skills/ - In Claude Code, invoke it by describing a fixture, e.g. "分析一下巴西对摩洛哥" or "今天北京时间6点那场世界杯怎么看". The skill confirms the fixture/timezone, gathers data, calls the trained engine for probabilities, and writes a Chinese report (and, on request, a mobile-first HTML page).
- The skill calls the prediction engine automatically; you can also run it standalone:
node worldcup-deep-analysis/tools/predict.mjs --rankHome 9 --rankAway 82 --host neutral # optional factors: --alt <m> --kaHome/--kaAway <n> --formHome/--formAway <x> - Validate any analysis JSON against the contract (zero dependencies):
npm run validate # checks samples/brazil-morocco.json node tools/validate-analysis.mjs <your-analysis.json>
See samples/ for a complete worked example (.json, .report.md, .html).
worldcup-deep-analysis/
SKILL.md skill entry (workflow + rules)
references/ dimensions, scoreline-math, prediction-model, sources, disclaimer
templates/ Markdown + mobile HTML report templates
tools/predict.mjs trained v6 Elo multi-factor engine (zero-dep, runnable)
schema/analysis.schema.json analysis-JSON contract (draft-07)
tools/validate-analysis.mjs zero-dependency validator
samples/ Brazil vs Morocco worked example
- Schedule / timezone (selectable) & qualification scenarios
- Squads — predicted/confirmed XI, bench, key players, season form
- Absences — injuries, suspensions, fitness doubts
- Coaches — style, formation, stability, in-match adjustments (inferred, labelled)
- Referee crew — tendencies (cards/penalties/added time) & psychological read (inferred)
- Venue & environment — pitch, altitude, weather, crowd lean, travel
- Head-to-head, market odds, and a calibrated probability model powered by the trained v6 Elo multi-factor engine (1X2, scoreline distribution, over/under, BTTS, xG, scenarios)
Hard data is cited with sources & timestamps; model/LLM inferences are explicitly labelled.
A full World Cup match-analysis web app built on this project is deployed and open-source:
- Live demo: wc.netzfs.com
- Website source: github.com/netz888/WORLD-CUP-AI-HUB (Next.js)
This project is a statistical & AI-assisted analysis tool for entertainment and discussion only. Predictions carry inherent uncertainty; historical accuracy does not guarantee future results. Not gambling advice.
一个用于世界杯比赛深度分析的 Claude Code skill：阵容、伤停、教练、裁判倾向、 球场/海拔/天气，以及经过校准的胜平负与比分概率。
仅供娱乐，非投注建议。
子项目 1（自包含 skill）已实现。 skill 本体、参考文件、模板、「分析 JSON」schema、
训练好的 Elo 多因子预测引擎（tools/predict.mjs）、零依赖校验器、以及一份完整的
巴西vs摩洛哥范例都已在仓库中。设计文档见 docs/superpowers/specs/。
概率来自 v6 Elo 多因子引擎（worldcup-deep-analysis/tools/predict.mjs），在
2010 / 2014 / 2018 / 2022 四届世界杯小组赛共 192 场真实赛果上做网格搜索 + 留一届交叉验证
训练而成。训练集 vs 验证集精确比分命中 ~16.1% vs ~16.7%（未过拟合）；1X2 方向命中 ~55%。
精确比分命中存在物理上限（~16–17%），引擎对此诚实标注。详见
references/prediction-model.md。
skill 自包含 —— 无需付费 API、无密钥。 Node 用于运行训练好的预测引擎与校验器。
- 把 skill 复制到你的 Claude Code skills 目录：
cp -r worldcup-deep-analysis ~/.claude/skills/ - 在 Claude Code 里用自然语言触发，例如："分析一下巴西对摩洛哥"、 "今天北京时间6点那场世界杯怎么看"。skill 会确认对阵/时区、搜集数据、调用训练好的引擎 算概率，产出中文报告（并可按需生成移动端 HTML 页面）。
- skill 会自动调用预测引擎；你也可以单独运行：
node worldcup-deep-analysis/tools/predict.mjs --rankHome 9 --rankAway 82 --host neutral # 可选因子：--alt <海拔米> --kaHome/--kaAway <核心缺阵数> --formHome/--formAway <近期净胜球趋势> - 用零依赖校验器检查分析 JSON 是否符合契约：
npm run validate # 校验 samples/brazil-morocco.json node tools/validate-analysis.mjs <你的分析.json>
完整范例见 samples/（含 .json / .report.md / .html）。
给定一场比赛（如"巴西 vs 摩洛哥"，或"今天北京时间 6 点那场"），skill 会先确认对阵与时区， 逐维度搜集数据、估算概率，输出深度分析报告（+ 可选移动端 HTML）：
- 赛程 / 时区（可选）与出线形势
- 阵容 —— 预测/确认首发、替补、核心球员、赛季状态
- 伤停 —— 伤病、停赛、出场存疑
- 教练 —— 风格、阵型、稳定性、临场调整 （推断，会标注）
- 裁判组 —— 倾向（场均牌/点球/补时）与"心理"解读 （推断）
- 场地与环境 —— 草皮、海拔、天气、球迷倾向、旅行
- 历史交锋、市场赔率、以及由训练好的 v6 Elo 多因子引擎驱动的校准概率模型 （1X2、比分分布、大小球、BTTS、xG、情景推演）
硬数据标注来源与时间；模型/LLM 的推断会明确标注，不伪装成数据。
基于本项目构建的完整世界杯赛事分析网站已部署上线、源码开源：
- 在线演示： wc.netzfs.com
- 网站源码： github.com/netz888/WORLD-CUP-AI-HUB（Next.js）
本项目为统计与 AI 辅助的赛事分析工具，仅供娱乐与讨论；预测存在固有不确定性， 历史准确率不代表未来结果。非赌博建议。