Skip to content

analyze.md

Latest commit

 

History

History
625 lines (461 loc) · 26.1 KB

analyze.md

File metadata and controls

625 lines (461 loc) · 26.1 KB
Raw

Workflow: Chart Analysis (with auto-annotation)

For when the user attaches a trading chart image and asks anything about it.

When this workflow applies

  • User attaches a chart image (TradingView / MT4 / Binance / 火币 / OKX / Bybit / any exchange UI screenshot)
  • User asks any of: analysis / patterns / where to enter / setup / structure / bias / risk
  • Chinese triggers: "分析" / "看图" / "看一下" / "帮我看" / "入场" / "标一下" / "形态" / "走势" / "做多" / "做空" / "止损放哪"

When NOT to use

  • Pure conceptual Q&A without a chart → use qna.md
  • Non-trading images (memes, code screenshots) → not this skill
  • User explicitly wants only annotation, no analysis → use annotate.md

Mandatory Workflow (7 steps)

Step 1: Examine the chart

Output a brief inventory before analysis:

Field How to read If unreadable
Asset Top-left ticker label (BTCUSDT / NQ1! / EURUSD / AAPL) mark null
Timeframe Top toolbar (1H / 4H / 1D / 15m) mark null
Visible price range Y-axis high/low mark null
Time range X-axis start/end mark null
Current price Right-side highlighted label on last candle mark null
Existing annotations Lines, boxes, fib levels, text labels already drawn list them
Candle pattern features Long wicks, displacement, consolidation, gaps objective description only

Critical: prefer null over guessing. Wrong prices poison the entire analysis.

Step 1a: Detect multi-panel images

Check whether the image contains multiple chart panels (e.g. 5m + 15m side-by-side, or 2x2 multi-timeframe).

  1. Count distinct panels — separate chart boxes with their own x/y axes and ticker labels
  2. Identify each panel's role:
    • Same asset, different timeframes → multi-timeframe analysis (HTF/LTF alignment)
    • Different assets → inter-market / correlation (e.g. SMT divergence)
    • Same asset, same timeframe, different overlays → comparative annotation view
  3. Read Step 1 fields for each panel separately, label them panel_left / panel_right etc.
  4. Note panel relationships explicitly ("panel_left is 5m, panel_right is 15m, same asset ETHUSDT — multi-timeframe view")

When using multi-panel:

  • HTF panel → bias + PD Array zones
  • LTF panel → entry trigger (CISD / MSS / FVG)
  • Confirm alignment before suggesting entry; if HTF and LTF conflict, downgrade confidence and explain

Step 1b: Assess chart resolution

Quality Behavior
High (single panel, ≥1200px, clear y-axis) Read exact prices; can claim "FVG at 73,250 ~ 74,180"
Medium (single panel, blurry y-axis OR multi-panel high-res) Round to nearest 10/100/major level; add "approximate" caveat
Low (small / multi-panel low-res / dense candles) Use relative descriptions ("near prior swing high"); downgrade overall confidence to low; add to missing_information: "high-resolution single-panel chart"

Critical: if resolution prevents precise reading, DO NOT fabricate exact prices. Use approximate or relative language.

Step 1d: Identify asset (for real-data fetch)

Try to extract from the chart:

Field How to read
Symbol / ticker Top-left label (e.g. "BINANCE:BTCUSDT.P", "TSLA", "600519")
Exchange Often part of the prefix (BINANCE:, BYBIT:, OKX:, NASDAQ:)
Market Spot vs perpetual is hinted by suffix (.P / PERP)
Interval Top toolbar selector ("1h", "4H", "1D", "5m")

Map to Mobius API canonical form:

Chart hint Mobius canonical
BINANCE:BTCUSDT.P binance / perp / BTCUSDT
BINANCE:ETHUSDT (no .P) binance / spot / ETHUSDT
BYBIT:BTCUSDT.P bybit / perp / BTCUSDT
SH:600519 / 贵州茅台 stock / cn / 600519
HKEX:00700 / 腾讯控股 stock / hk / 00700
NASDAQ:AAPL stock / us / AAPL

If you can identify symbol but not the exchange/market explicitly, use:

.venv/bin/python scripts/kb_klines.py resolve "<natural name or ticker>"

If asset identification fails (resolution low, no visible ticker, unrecognized) → skip Step 1e, proceed with visual-only analysis, and add "high-resolution single-panel chart" / "clearly visible ticker label" to missing_information.

Step 1e: Fetch real OHLCV (data-grounded analysis)

If Step 1d succeeded, fetch real data:

.venv/bin/python scripts/kb_klines.py fetch \
    --exchange <ex> --market <mkt> --symbol <sym> \
    --interval <tf> --limit 200 \
    --with-htf \
    --output <image_dir>/<image_stem>.klines.json

Step 1f: Fetch SMC structural indicator (default)

The SMC indicator gives a structural ground-truth reading to complement the visual chart analysis:

.venv/bin/python scripts/kb_klines.py indicators \
    --exchange <ex> --market <mkt> --symbol <sym> \
    --interval <tf> --limit 200 --format compact \
    --output <image_dir>/<image_stem>.smc.txt

Read the .smc.txt with the Read tool. It contains exact prices for:

  • Per-bar trend bias (swing & internal)
  • Active swing / internal pivots, trailing extremes with Strong/Weak labels
  • objects sidecar: Order Blocks (active/mitigated), Fair Value Gaps, equal highs/lows, premium/equilibrium/discount zones, BOS/CHoCH events
  • alerts_last_bar: structural events that fired on the latest candle

Use the field-semantics map in SKILL.body.md to consume each section.

Step 1g: Verify freshness

The .smc.txt header now includes (or the JSON response has) a freshness block:

data_source:          Mobius Quant API (api.mobiusquant.ai)
fetched_at (UTC):     2026-05-23T15:21:14Z
last_bar_open (UTC):  2026-05-23T15:00:00Z
last_bar_age:         1274s (interval=3600s, is_stale=False)
current_price:        75609.9

You MUST carry these values verbatim into the Step 6 output footer. Never answer using prices, structures, or pivots from training data or older conversation turns — Step 1e/1f re-grounds you in fresh data.

If is_stale=True, prepend a warning line to the final reply.

Fallback to local extraction (only if SMC API unreachable):

.venv/bin/python scripts/kb_klines.py analyze \
    --input <image_dir>/<image_stem>.klines.json \
    --output <image_dir>/<image_stem>.features.txt

Sanity check — compare data to chart:

Check Action
API current_price vs chart's last close (if readable) If within 2% → OK; if > 2% → chart is likely historical, warn user and ask for time hint
API timeframe range vs chart's visible range If wildly different → image likely shows different period than fetched; flag in missing_information
Asset on chart matches API resolve If user-provided symbol differs from chart label → tell user

Opt-out: if the user said "只看图" / "skip data" / "no API" → skip Step 1e entirely; mark data_source: "visual_only" in JSON.

Fetch failure (network error, symbol not on Mobius, 429 rate limit max retries) → log the failure, proceed with visual-only, add data_fetch_failed: <reason> to missing_information.

Step 1c: Calibrate coordinate system (required for auto-annotation)

For EACH panel:

chart_bbox (pixel coordinates of plotting area):

  • x: left edge of price grid (right of toolbar — NOT image edge)
  • y: top edge of plotting area (below title/menu bar)
  • width: from left grid edge to right edge of price grid (exclude right-side price labels if outside grid)
  • height: top to bottom of grid (exclude bottom time-scale and indicator panels)

y_axis_range:

  • top: highest price label
  • bottom: lowest price label

theme: "dark" or "light" (based on chart background)

Why: Step 7 (auto-annotation) uses these to convert price → pixel position. Wrong bbox → annotations drawn outside chart.

Best practices:

  • Use image dimensions (e.g. Read tool returns size) to bound estimates
  • Conservatively estimate (smaller bbox > overshoot)
  • If unsure → set chart_bbox: null and y_axis_range: null in JSON; Step 7 will skip

Step 2: Form preliminary hypotheses

Candidate patterns from this set:

  • Patterns: FVG / Order Block (OB) / Breaker Block / Mitigation Block / Inversion FVG (IFVG)
  • Liquidity: Liquidity Sweep / Buy-Side Liquidity / Sell-Side Liquidity / Inducement / Stop Run
  • Structure: BOS / CHoCH / MSS / Higher High / Lower Low
  • Confirmation: CISD / Displacement / Imbalance / Volume Imbalance
  • Zones: Premium/Discount / OTE (0.62-0.79) / Mean Threshold / Equilibrium
  • Timing: Killzone / London Session / NY Open / Asia Range / Power of 3 / Silver Bullet

If a pattern looks possible but you're unsure, still retrieve — let the cards decide.

Step 3: Retrieve concepts from the knowledge base

.venv/bin/python scripts/kb_retrieve.py "<keywords>" --top-k 5

Examples:

# Generic chart with long wicks
kb_retrieve.py "long lower wick liquidity sweep reversal" --top-k 5

# FVG + entry
kb_retrieve.py "Fair Value Gap entry OTE CISD confirmation" --top-k 5

# Similar historical case
kb_retrieve.py "BTC 4H liquidity sweep entry reversal" --type case --top-k 3

# Single school
kb_retrieve.py "smart money concepts market structure" --school ICT --top-k 5

Multiple retrievals are encouraged for complex charts.

Step 4: Apply rules to chart + data (dual-source)

For each retrieved card:

  1. Read its identification_rules
  2. Match each rule against two evidence sources (when both available):
    • Visual evidence from the chart (candle features, structure events)
    • Data evidence from the features.txt produced by Step 1e (exact prices, mitigation %, sweep wicks, displacement strength)
  3. Confirm the pattern only if at least one source clearly satisfies the rule. Strong confirmation = both sources agree.
  4. Reject if both sources fail the rule.
  5. Note common_mistakes to avoid.

Data takes precedence on price levels: when the user asks "where's the FVG", quote the exact range from features.txt, NOT a visual estimate.

Visual takes precedence on subjective features: chart annotations / drawn lines / user notes only exist in the image.

Conflict handling: if features.txt says "no FVG at level X" but the chart visually looks like there might be one, you likely misread the chart. Trust data; mention the conflict in the Analysis section.

Citation format:

  • Visual: "Rule 2 of FVG: 'high of first candle below low of third' — visible at candles 12:00 / 16:00 / 20:00 forming gap"
  • Data-grounded: "Bullish FVG confirmed at 73,182 - 74,210 (features.txt), 33% mitigated. Rule 2 satisfied: c0.high (73182) < c2.low (74210)."

Reject example: "Rejected Order Block hypothesis — features.txt shows no displacement > 1.5× ATR in next 3 candles; Rule 3 fails"

Step 5: Save structured JSON (silent, NOT shown to user)

Save to: <image_directory>/<image_stem>.analysis.json

Example: /home/Codes/QuantKnowledge/test-charts/eth_5m.png → /home/Codes/QuantKnowledge/test-charts/eth_5m.analysis.json

Use the Write tool. DO NOT paste the JSON content in your reply. It's for downstream tools.

JSON schema:

{
  "input_image": "<absolute path of original image>",
  "asset": "<ticker or null>",
  "timeframe": "<e.g. 4H or null>",
  "visible_price_range": [<low>, <high>] | null,
  "current_price": <number or null>,

  "data_source": "visual+api" | "visual_only" | "api_only",
  "klines_json_path": "<path to .klines.json if Step 1e succeeded>" | null,
  "features_path": "<path to .features.txt if Step 1e succeeded>" | null,
  "data_chart_consistency": "match" | "mismatch_warn" | "n/a",

  "chart_bbox": {"x": <int>, "y": <int>, "width": <int>, "height": <int>} | null,
  "y_axis_range": {"top": <number>, "bottom": <number>} | null,
  "theme": "dark" | "light",

  "trend": "bullish" | "bearish" | "consolidating" | "uncertain",
  "bias": "long" | "short" | "neutral" | "uncertain",
  "patterns": [
    {
      "type": "FVG",
      "range": [<low>, <high>],
      "label": "<short label>",
      "confidence": "very_high" | "high" | "medium" | "low" | "very_low",
      "source_card": "<card id from retrieval>"
    }
  ],
  "trade_setup": {
    "entry": {"price": <number>, "label": "<rationale tag>"} | null,
    "stop_loss": {"price": <number>, "label": "<...>"} | null,
    "targets": [{"price": <number>, "label": "<T1: prior high>"}]
  },
  "outcome_cases": [
    {
      "case_id": "A",
      "probability": "very_high" | "high" | "medium" | "low" | "very_low",
      "scenario": "<one-line description>",
      "trigger_signals": ["<signal1>", "<signal2>"],
      "action": "<what to do>",
      "invalidation": "<what kills this case>"
    }
  ],
  "risks": ["<risk1>", "<risk2>"],
  "confidence": "very_high" | "high" | "medium" | "low" | "very_low",
  "missing_information": ["<what's unclear>"]
}

JSON rules:

  • null for any field you cannot determine
  • Every price must appear on the chart or derive from a retrieved rule
  • If too uncertain for 2+ outcome_cases, leave outcome_cases: []
  • chart_bbox / y_axis_range null → Step 7 skipped

Step 7: Generate chart image — two options

You have two paths to produce a chart image. Pick based on user intent and image quality.

Option A: Annotate user's uploaded image (PIL)

Use when the user wants markup on their own chart (their TradingView screenshot, existing drawings, etc.).

.venv/bin/python scripts/kb_phase_b_to_c.py \
    --input <image_dir>/<image_stem>.analysis.json \
    --image <original_image_path> \
    --output <image_dir>/<image_stem>.annotated.png

Reads chart_bbox / y_axis_range / theme from JSON, maps patterns + trade_setup to annotations, renders on top of the original image.

Skip Option A if:

  • chart_bbox is null (couldn't calibrate)
  • Multi-panel image (current version supports single-panel annotation only)

Option B: Generate fresh TradingView-grade chart (lightweight-charts)

Use when:

  • User asked "出张图" / "画张干净的图" / "重新画一张"
  • User's image is low-resolution (Step 1b judged low)
  • Step 1e fetched real data (Option B is more accurate since it uses the SAME data)
# 1. Pull pure K-lines → panels payload skeleton (items=[])
.venv/bin/python scripts/kb_klines.py chart \
    --exchange <ex> --market <mkt> --symbol <sym> --interval <tf> \
    --limit 200 \
    --output <image_dir>/<image_stem>.chart.json

kb_klines.py chart auto-fills the structural overlay from the SMC indicator — no manual panels[0].items authoring needed. If you have a trade-setup (entry / SL / target) to draw, write a small JSON file with only those hlines:

cat > <image_dir>/<image_stem>.setup.json <<'JSON'
{"items": [
  {"type": "hline", "value": 78500, "label": "Short 78500",
   "style": {"role": "entry_short", "width": 2}},
  {"type": "hline", "value": 80000, "label": "SL 80000",
   "style": {"role": "stop_loss", "dash": "dashed", "width": 2}},
  {"type": "hline", "value": 77000, "label": "T1 77000",
   "style": {"role": "target", "width": 2}}
]}
JSON

Render (pass --trade-setup only if you authored a setup file):

.venv/bin/python scripts/kb_klines.py render \
    --input <image_dir>/<image_stem>.chart.json \
    --trade-setup <image_dir>/<image_stem>.setup.json \
    --output <image_dir>/<image_stem>.chart.png \
    --theme dark --width 1400 --height 900

See workflows/klines.md Step 4 for the auto-overlay knobs (--max-items, --no-include-mitigated, etc.) and trade-setup label rules.

Skip both if

  • No actionable trade_setup AND no drawable patterns
  • Image is non-trading

Step 6: Output conversational reply (5 sections, user-facing)

Output format is MANDATORY. The reply MUST use the four section headings below verbatim, in this exact order, in the user's language:

  • ## 结论 / Conclusion
  • ## 分析逻辑 / Analysis
  • ## 后续走势与操作 / Outcome Cases
  • ## 风险与失效 / Risks & Invalidation

A fifth section ## 信息缺失 / Missing Information is optional and only added when confidence ≤ medium.

Free-form prose without these ## headings is an incomplete reply and must be rejected before sending to the user.

Reply structure (verbatim) in the user's language:

## 结论 / Conclusion
- **Bias**: <long-leaning / short-leaning / neutral / uncertain>
- **Confidence**: <very_high / high / medium / low / very_low>
- **操作建议 / Action**: <one-line concrete recommendation, e.g. "等 2245 retest 后做空，SL 2270，目标 2210/2200">
- **关键依据 / Key evidence (≤3)**: <bullet list of 2-3 most decisive signals>

## 分析逻辑 / Analysis

Walk through the causal chain **observation → knowledge-base rule → conclusion**:

1. **图上观察 / What's on the chart**: candle features, structural events, key levels
2. **知识库匹配 / Knowledge base hits**: which retrieved cards apply
3. **规则推导 / Rule application**: cite specific rules; show how chart features satisfy or violate them
4. **驳回的可能性 / Rejected hypotheses**: patterns that looked plausible but failed rule checks (with reason)

## 后续走势与操作 / Outcome Cases

List 2-3 plausible scenarios using the 5 probability tiers (`very_high` / `high` / `medium` / `low` / `very_low`).

For each case:
- **Case <letter> (<probability tier>)**: <scenario description>
- **触发信号 / Trigger signals**: <observable signals confirming this case>
- **操作建议 / Action**: <concrete entry/stop/target OR "观望 / wait for X">
- **失效条件 / Invalidation**: <what kills this case>

If chart is too ambiguous for 2+ scenarios, replace with:

> 当前结构信息不足以给出多场景预测：
> - 缺失：<list of missing info>
> - 建议：等 <specific signal> 后再评估

**Do NOT force scenarios when evidence is thin.**

## 风险与失效 / Risks & Invalidation

- **主要风险 / Main risks** (2-4 from retrieved cards' `common_mistakes`)
- **整体 setup 失效条件 / Overall invalidation**: <what would invalidate the entire bias>
- **监控提示 / Monitoring hints**: <e.g. "wait for NY Killzone before entry">

## 信息缺失 / Missing Information (optional, only if confidence ≤ medium)

- <list of missing pieces that would raise confidence>

After the 5 sections, append the mandatory freshness footer — values come directly from Step 1g's API response, do NOT fabricate them:

📅 数据时点 / Data as of (UTC): <freshness.last_bar_open_time_utc>
🕐 当前价 / Current price:     <current_price>
📡 数据源 / Source:            Mobius Quant API → <exchange>:<market>:<symbol> @ <interval>
🔍 拉取时刻 / Fetched at (UTC): <freshness.fetched_at>
⏱️  K 线年龄 / Bar age:        <freshness.last_bar_age_seconds>s (is_stale=<freshness.is_stale>)
📂 分析数据 / Analysis JSON:   <absolute path>
🖼️ 标注图 / Annotated chart:  <absolute path>  ← only if Step 7 succeeded

If freshness.is_stale == true, prepend a top-level warning line at the very start of the reply: ⚠️ 数据可能滞后 / Stale data warning: latest <interval> bar is <age>s old.

If Step 1e/1f was skipped (visual-only mode), omit the freshness lines and mark the reply with (visual-only — no live data fetched).

If Step 7 was skipped, omit the annotated chart line and optionally add: (标注图未生成：<reason, e.g. chart_bbox 无法可靠标定 / multi-panel 暂不支持>)

Constraints (re-emphasized)

  1. No fabrication (shared rule)
  2. No analytical adjectives without basis — avoid "strong support", "obvious reversal", "clear breakout" unless backed by a retrieved rule + visible evidence
  3. State uncertainty explicitly — prefer "uncertain — "
  4. Cite the knowledge base — every confirmed pattern references a retrieved card
  5. Multiple retrievals OK — complex charts may need 2-3 retrieves
  6. Language rules (shared) — Chinese prose / English technical terms
  7. Non-trading image — say "Not a trading chart" and stop
  8. Multi-panel — read each panel separately, label them, note relationship
  9. Resolution-aware — downgrade to approximate / relative when low res
  10. JSON is internal — silently save; NEVER paste in conversational reply
  11. Don't force outcome_cases — if too ambiguous, skip Cases and explain what's needed

Examples

Example 1 — ETH 5m clear bearish setup (Chinese user)

User: [attaches ETH 5m chart] "查看一下这个图片，给我分析一下当前的行情"

Internal workflow:

  1. Step 1: ETHUSDT.P 5m, image 1316×708, range 2200-2280, current 2226.5
  2. Step 1c: chart_bbox = {x:50, y:30, w:1186, h:510}, y_axis_range = {top:2280, bottom:2200}, theme="dark"
  3. Step 2: hypotheses → Liquidity Sweep at 2245, Bearish Displacement, Breaker Block
  4. Step 3: kb_retrieve.py "liquidity sweep breakdown displacement breaker 5m short setup"
  5. Step 4: confirm all three; reject V-shape rebound (multi-candle, not 1-3 candles)
  6. Step 5: silently save JSON to /home/Codes/QuantKnowledge/test-charts/eth_5m.analysis.json
  7. Step 7: run kb_phase_b_to_c.py → saves eth_5m.annotated.png
  8. Step 6: 5-section reply in Chinese

Reply (Chinese prose, English technical terms):

## 结论 / Conclusion
- **Bias**: short-leaning（短期）
- **Confidence**: medium
- **操作建议**: 等 2245 retest + LTF CISD 后做空，SL 2270，目标 2210 → 2200
- **关键依据**:
  - 21:00 sell-side Liquidity Sweep at 2245（横盘支撑被穿）
  - Bearish Displacement (巨量阴线 2245 → 2205)
  - 当前 2226.5 在 0.5 retracement 区，未明确突破 2245

## 分析逻辑 / Analysis

**图上观察**:
- 11:00-19:00 在 2245-2280 区间震荡
- 21:00 巨量阴线，2245 关键支撑被跌穿
- 22:30 触底 2205
- 当前 2226.5 反弹至 0.5 retracement

**知识库匹配**:
- Liquidity Sweep（28 视频源）: 横盘低被穿
- Displacement: 快速 expansion 收盘穿透
- Breaker Block: 跌破的 2245-2255 等待 retest 转阻力

**规则推导**:
- Liquidity Sweep Rule 1 ✓: 价格扩张跌破前低
- Displacement Rule ✓: 快速 expansion + 收盘穿透
- Breaker formation in progress: 2245-2255 待 retest 确认

**驳回**:
- ❌ V-Shape Reversal: 反弹跨多根 K 线，不符合 1-3 根快速反转标准
- ❌ FVG entry at 5m: 未明确 CISD 触发，不足以入场

## 后续走势与操作 / Outcome Cases

**Case A (high)**: 反弹至 2245 breaker 后受阻继续下跌
- **触发信号**: 1m/5m 在 2240-2245 形成 rejection wick + CISD
- **操作建议**: 空头入场 2240-2245, SL 2270, T1 2210, T2 2200
- **失效条件**: 5m 收盘上穿 2255

**Case B (medium)**: 突破 2245 后向上回测 2260-2270
- **触发信号**: 5m 收盘上穿 2245 + 量能放大
- **操作建议**: 暂停空头计划，等 2260-2270 zone 反弹结束再评估
- **失效条件**: 突破 2270

**Case C (low)**: 进入 2210-2230 range bound
- **触发信号**: 多次测试 2210/2230 双向不破，量能萎缩
- **操作建议**: 观望或 range scalp（多 2210，空 2230）
- **失效条件**: 任一侧带量突破

## 风险与失效 / Risks & Invalidation
- **主要风险**:
  - 反弹诱多陷阱（Liquidity Sweep common_mistakes #2）
  - 未等 LTF CISD 就入场（Breaker Block 规则要求）
- **整体 setup 失效**: 5m 收盘上破 2270
- **监控**: 等 5m K 线接近 2245 时的反应（rejection 还是 break-through）

📂 分析数据 / Analysis JSON: `/home/Codes/QuantKnowledge/test-charts/eth_5m.analysis.json`
🖼️ 标注图 / Annotated chart: `/home/Codes/QuantKnowledge/test-charts/eth_5m.annotated.png`

Example 2 — English user, ambiguous chart

User: [attaches BTC 1D long-range chart] "What's happening here?"

Reply (English, technical terms kept):

## Conclusion
- **Bias**: uncertain
- **Confidence**: low
- **Action**: No actionable setup; wait for LTF (4H/1H) entry triggers
- **Key evidence**: Macro discount zone + VWAP overhead + mixed short/medium term momentum

## Analysis

**Chart observations**: BTC/USDT 1D, range 60K-105K, current 79.4K, VWAP 80.4K overhead.
30D +6%, 180D -15%.

**Knowledge base hits**: V-Shaped Reversal (macro form only), Premium/Discount Array, HTF/LTF Analysis.

**Rule application**: Price below EQ (82.5K) → in discount zone (long-favorable per Premium/Discount Array). V-Shape Rule 2 (1-3 candle reversal) fails for this multi-month rebound — caveat: "macro V" only.

**Rejected hypotheses**:
- ❌ FVG: daily timeframe resolution insufficient for 3-candle pattern
- ❌ Order Block: no CISD displacement visible
- ❌ Liquidity Sweep at 60K: chart doesn't show pre-Nov-2025 history to confirm

## Outcome Cases

当前结构信息不足以给出多场景预测：
- Missing: LTF (4H / 1H) charts for entry triggers + HTF (Weekly/Monthly) for macro bias
- Recommended: Drop to 4H/1H, look for CISD or MSS before considering directional bias

(Not forcing scenarios — chart genuinely too ambiguous at daily-only view.)

## Risks & Invalidation
- **Main risks**: Treating multi-month rebound as canonical V-Shape; anchoring stop to 60K macro low (poor R/R for daytrade)
- **Setup invalidation**: N/A (no setup to invalidate)

## Missing Information
- LTF (4H / 1H) for CISD / FVG / OB entry triggers
- Weekly / Monthly for macro bias
- Pre-Nov-2025 history to confirm 60K as macro swing low

📂 Analysis JSON: `/home/Codes/QuantKnowledge/test-charts/btc_1d.analysis.json`
(标注图未生成：无可执行的 trade_setup，跳过自动 annotation)

Example 3 — Non-chart image

User: [attaches a random meme image] "分析一下"

Reply (single line): "Not a trading chart. Please upload a candlestick / line / OHLC chart for analysis."

No JSON, no 5-section output.

Example 4 — Multi-panel multi-timeframe

User: [attaches 5m + 15m ETHUSDT side-by-side] "帮我看一下"

Internal: detect 2 panels (panel_left=5m, panel_right=15m), assess medium-low resolution, retrieve HTF/LTF concepts, build alignment-based analysis.

Reply: same 5-section format. In Analysis section, note multi-panel + HTF/LTF alignment. In Cases, scenarios reference both timeframes. In Missing Info, mention "high-resolution single-panel HTF chart" if resolution is bottleneck. Step 7 will be skipped (multi-panel annotation not yet supported); add the "(标注图未生成)" note.