feat: per-ticker fear-and-greed reading on replay rows

[yanivil]

Description

A per-ticker fear-and-greed reading, recorded as a replay feature under the protocol of docs/wiki/03 and tested over the eleven yearly point-in-time runs from this branch. No rule changes.

• fear_greed: the equal-weight 0-100 composite of the three price-based oscillators the TradingView community indicators of that name share: RSI 14 (Wilder's smoothing, _rsi), the MACD (12, 26, 9) histogram as its mid-rank percentile within the trailing 250 bars, and Bollinger %B over 20 bars and 2 standard deviations, clipped to 0-100 in the average.
• Two readings per row: the composite and its components at the scan day (fg_score, fg_rsi, fg_macd_pct, fg_bb_pctb) and the composite at the pattern's last low (fg_base).
• Buckets in the outcome-by-feature table; wiki 03 "Per-ticker fear and greed, tested"; wiki 04 feature list; changelog.

Results (runs 34267945406 to 34268343900, tuned, 2016 to 2026):

Reading	Zone	N	Hit	Mean R	Positive years
scan day	fear	73	23 %	+0.61	6 of 11
scan day	neutral	174	40 %	+0.45	10 of 11
scan day	greed	957	33 %	+0.20	10 of 11
scan day	extreme greed	427	42 %	+0.12	8 of 11
pattern's last low	extreme fear	112	35 %	+0.30	7 of 11
pattern's last low	fear	352	34 %	+0.28	8 of 11
pattern's last low	neutral	616	39 %	+0.24	10 of 11
pattern's last low	greed	534	33 %	+0.15	8 of 11

The greedier the stock at its breakout, the lower the mean R; bases formed in fear paid about twice what bases formed in greed did. The stretch components carry it: RSI 30-50 at the breakout +0.55 R against +0.12 above 70; Bollinger %B lower half +0.53, upper half +0.26, a close above the upper band +0.04 R on 453 signals, the upper half ahead in 10 of 11 years. Momentum (MACD percentile) does not: +0.18 to +0.26 across buckets. Leaving out above-band breakouts would keep 1179 signals at +0.30 R against 1632 at +0.22, at a cost of 16 R of the ten-year 367. Still a hypothesis under the protocol: the gate has to be replayed as a rule on its own for its drawdown and per-pattern effect.

Verified: ruff clean, 137 tests pass (the composite on accelerating, collapsing, flat and too-short series, Wilder's RSI by hand, %B outside the bands, the cup fixture's readings recomputed independently with the base more fearful than the breakout, rows and buckets rendered); a 30-ticker real-data run renders the buckets.

Related Issues

Related: #99 and #101 (stock-level and market-level context), #108 (drawdowns), #96 (protocol)

Checklist

• Code follows existing project style and type annotations
• Threshold changes include a one-line "why" comment in scan.py (no thresholds changed)
• Offline unit tests pass (python -m pytest -q)
• Random-walk false-positive rate stays under 5% (detection logic untouched)
• Documentation updated in docs/wiki/
• CHANGELOG.md updated

🤖 Generated with Claude Code