BTC-first → 10-asset portfolio → out-of-sample. Conservative costs (~0.21% round-trip). Frozen config: 5m, rr×1.0, stop 4.0 ATR.
The lower-TF edge is real and survives out-of-sample (5m OOS PF 1.418, t=2.49 — degrades gracefully from IS, not overfit) and it takes +19% more trades, the original goal. But it does not beat V5: the 15m baseline holds a higher PF (1.825 vs 1.418) at roughly half the drawdown, both IS and OOS. The binding constraint is edge-density/correlation of the pullback entry — not the timeframe. Recommendation: keep V5; pursue a second regime-orthogonal 15m entry + correlation gate (V6), not a naive lower-TF port.
| Run | Trades | Win% | PF | Return | MaxDD | t-stat |
|---|---|---|---|---|---|---|
| IN-SAMPLE (Binance, 2020 → 2024-04) | ||||||
| 15m baseline (V5) | 600 | 62.7% | 1.825 | 218.9% | -11.44% | 5.11 |
| 5m locked | 711 | 60.3% | 1.535 | 191.2% | -18.35% | 4.12 |
| OUT-OF-SAMPLE (Binance, 2024-04 → 2026-05) — frozen config, run once | ||||||
| 15m baseline (V5) | 315 | 61.3% | 1.825 | 60.4% | -8.07% | 3.79 |
| 5m locked | 376 | 59.0% | 1.418 | 43.5% | -10.64% | 2.49 |