72_candlestick_scanner

72 — Candlestick Pattern Scanner

Category: Technical Analysis / Screening

Pure-Python recognition of 9 classic candlestick patterns on live K-line data. No ML, no external deps — just geometry and heuristics.

Patterns Detected

Pattern	Bars	Signal
⚖️ Doji	1	Indecision (bias from shadow direction)
🔨 Hammer	1	Bullish reversal (after downtrend)
🌠 Shooting Star	1	Bearish reversal (after uptrend)
🟢 Bullish Engulfing	2	Strong bullish reversal
🔴 Bearish Engulfing	2	Strong bearish reversal
🌅 Morning Star	3	Bullish reversal
🌆 Evening Star	3	Bearish reversal
📈 Three White Soldiers	3	Bullish continuation
📉 Three Black Crows	3	Bearish continuation

How It Works

1. Fetches historical K-lines via request_history_kline for warmup.
2. Subscribes to CUR_KLINE push for real-time bar completion.
3. On each new bar, runs all 9 pattern detectors on the trailing window.
4. Confidence scoring combines pattern strictness + trend alignment + volume.
5. Cooldown timer prevents duplicate alerts (configurable, default 5 bars).

Confidence Scoring

• Base: Pattern geometric strictness (0–1)
• Trend bonus: +0.15 max if pattern aligns with rolling slope
• Trend penalty: −0.10 if pattern contradicts trend
• Final confidence clamped to [0%, 100%]

Usage

python3 main.py                         # defaults to HK.00700
python3 main.py --stock HK.00700 --lookback 30

Key Concepts Demonstrated

• 9 pattern detectors as pure functions
• CurKlineHandlerBase push subscription
• Confidence scoring with trend confirmation
• Cooldown-based alert deduplication
• Pure stdlib (no TA-Lib, no sklearn)