MOO-955 — 5 AUDITS Final Verdict (Apr 26)
5 parallel audits с правильной методологией · per-trade+daily Sh · bootstrap CI · WF OOS · 5-10bps slippage · supersedes Apr 26 daily-Sh recheck
SUMMARY
Audit-1 M11 stack
2× WORSE
baseline Sh 6.25 vs M11 3.00
Audit-2 BREAK_UP
VALIDATED
all tiers reproduce CI 95%
Audit-3 PEAD
B FAILS
A stable Sh 2.79, drop B
Audit-4 Look-ahead
9 GREEN
2 yellow (IMB fix), 0 red active
Audit-5 Regime
F NARROW
D drop, 2026=noise
Critical: 2026 M11 sh_daily −1.24 — real recent breakdown, не bootstrap noise. WF OOS 2024-25 train → 2026 test confirms over-fit. Apr 26 daily-Sh recheck N-biased (high-N strategies inflate), но conclusion partially correct — D4/Q/C1M действительно вредят.
FINAL FACTOR VERDICT
| Factor | Source | Verdict | Δ Sh / Impact | Action |
|---|
| EXIT 11:00 (vs 09:55) | A4 GREEN | KEEP | top contributor | — |
| QQQ hedge 1:1 | A4 GREEN, A1 | KEEP | +0.42 Sh_pt | — |
| B drop_BREAK_DN_LONG | A4 GREEN, A1 | KEEP | ~0 (NOISE) | — |
| BREAK_UP S+/S/A/B tiers | A2 reproduced | KEEP | Sh 9.6–19.2 daily | S+ floor 0→0.001 |
| Factor E (BMO clean) | A3 28/28 verified | KEEP | Sh 8.56 | — |
| noSHORT_S12 | A1 NOISE | KEEP | ~0 | cleanup OK |
| noBREAKUP_AVOID | A1 NOISE | KEEP | ~0 | — |
| noRsi30_LONG_forbid | A1 NOISE | KEEP | ~0 | — |
| F_no_macro (broad skip) | A5 BREAKING | NARROW | F-only OOS 10.16 > F+all 9.05 | skip PPI only, drop FOMC/CPI/NFP/OPEX skip |
| D4_sector_top4 | A1 +1.43 LOO vs A5 OOS DROP | DOWNGRADE | 2024 TOP4 8.83 < BOT11 9.62; 2026 inversion | switch to "drop rank=11 only" |
| IMB_SIZE_GATE | A4 YELLOW | FIX | +0.50 Sh after fix | rolling 60d median rewrite |
| C1M_dvol_ge_1M | A1 HARMFUL | REMOVE | −1.00 Sh_pt, −$11K/yr | delete |
| Q_laggard_LONG | A1 HARMFUL | REMOVE | −2.69 Sh_pt, −$11.5K/yr | delete |
| P_twist_top25_LONG | A1 NOISE | REMOVE | ~0 (cleanup) | delete (simplify) |
| H_short_float | A4 RED snapshot | ALREADY OFF | n/a | — |
PEAD BROKER (audit-3)
Tier S
N=9
collapsed with reconstructed scoring → MONITOR (was N=29 strict cutoff)
Tier A
VALIDATED
N=479 WR 59.7% [55.3, 64.1] Sh 2.79 stable 2024 (3.45) / 2025 (2.88) / 2026 (2.51)
Tier B FAILS honest revalidation: N=120 WR 45.8% Sh −2.76. Negative every year (2024 −3.11, 2025 −1.94, 2026 −4.43). Original Apr 23 backtest claimed Sh 2.23 — was bias.
Action: disable Tier B in broker/strategies/earn_pead_module.py.
METHODOLOGY DELTA vs Apr 26 RECHECK
| Issue | Apr 26 recheck | 5 audits (correct) |
|---|
| Sharpe metric | daily-aggregated only (N-biased) | per-trade + daily side-by-side |
| Slippage | not applied | 5-10bps roundtrip |
| Capacity normalization | $/yr = sum × $5K (raw) | per-position-deployed-daily |
| Stack search | greedy on noise (Sh 14.14→14.17) | LOO + WF OOS |
| Per-tier CI | none | bootstrap 1000 resamples 95% |
| 2026 claim | "strengthening Sh 20.70" | noise (66 days, CI overlaps prior years) BUT actual M11 2026 NEGATIVE −1.24 |
RISK FLAGS
- 2026 M11 sh_daily −1.24 NEGATIVE — recent breakdown confirmed, not bootstrap noise
- HIGH_VOL regime Sh 7.65 [CI 4.23] vs LOW_VOL 10.59 — calm-market biased → add VIX gate for sizing
- 2026 N=66 days — "strengthening" AND "breakdown" both small-sample, monitor 2 weeks
- S+ N=30 across 3y — regime-shift sensitivity for narrow tiers
- marketbeat fresh (Apr 26 01:26) — Apr 23 stale flag resolved
RECOMMENDED M11_v2 (LIGHTER)
Apply before next live day (2026-04-27 / next session):
1. strategies/moo_955/m10r_rules.py — comment out Q_laggard, C1M_dvol_ge_1M, P_twist filters
2. strategies/moo_955/config.json — apr25.boosts.dvol_filter=false, sector_filter→{drop_rank_11_only}, skip_macro→{PPI only}
3. broker/strategies/earn_pead_module.py — disable Tier B
4. imbalance_signal_backtest.py:97 — IMB rolling 60d median
5. strategies/moo_955/scorer.py — BREAK_UP S+ mag_intraday floor 0→0.001
KEEP unchanged: EXIT_1100, QQQ hedge, B drop_BREAK_DN_LONG, BREAK_UP S/A/B tiers, E BMO clean, noSHORT_S12, noBREAKUP_AVOID, noRsi30, PEAD A
FILES
| Audits (read-only) | revalidation_apr26/audit{1-5}_*.{py,csv,md} |
| Memory | research_moo955_5audits_apr26.md (top of MEMORY.md) |
| Apr 26 recheck (superseded partial) | research_moo955_recheck_1100_apr26.md |
| M11 deploy notes | project_moo955_M11_DEPLOYED_apr26.md |