1.06M M1 candles · walk-forward · barrier math · 1:4 RR · audited 2026-10-12
| Family | Pure WR (base 20%) | Hybrid WR (base 50%) | E[R] |
|---|---|---|---|
| FP-pullback + trend | 19–21% | 46–51% | ≤ 0 |
| HTF drift + deep pullback | 20–23% | 46–52% | ≤ 0 |
| Momentum ignition | 18–20% | 46–52% | ≤ 0 |
| Stat | Leaked (INVALID, retracted) | Leak-free (truth) |
|---|---|---|
| Model separation E[p|TP] vs E[p|SL] | 0.245 vs 0.195 | 0.220 vs 0.219 (zero) |
| Hybrid WR @ thr=0.25 | 58.3% | 49.0% (= baseline) |
| E[R] @ thr=0.25 hybrid | +0.13R | -0.16R |
P(+4R before −1R | BM) = 1/5 = 20%
Hybrid trigger P(+1R before −1R) = 50%
Labeler vs theory on synthetic RW: 0.198–0.200 vs 0.200 · 100% agreement with manual scan
EV gate: p·4 − (1−p) − fee > 0
✅ Data integrity (1,059,978 rows, 0 dups, 0 OHLC violations)
✅ Labeler = theory + manual scan
✅ Engine unit tests (fills, fees, ties)
✅ Feature causality after fix (truncate test)
🔍 LOOKAHEAD LEAK found in day_range_pos → FIXED