6 of 7 blocking checks failed — judged on ETH-USD 1h bars from coinbase, 2024-01-02 → 2026-05-01, with the family's ledger at N = 2 trials.
| State | Check | Value | Hurdle | Detail |
|---|---|---|---|---|
| FAIL | Deflated Sharpe | 0.645 | ≥ 0.95 | deflated Sharpe against N=2 trials |
| FAIL | Probability of backtest overfitting | 0.7 | < 0.5 | combinatorially symmetric CV |
| FAIL | t-statistic · Harvey–Liu hurdle | 0.49 | > 3 | Harvey-Liu hurdle, not the conventional 2 |
| FAIL | Backtest length · years | 1.413 | ≥ 138.4 | MinBTL = 2·ln(2) / E[max]² |
| FAIL | Regime survival · regimes that lost money | bear, chop | none | tested: ['bear', 'bull', 'chop'] |
| PASS | Net expectancy · bps per trade | 85.15 | > 0 | after the pinned cost model |
| FAIL | Trades | 5 | ≥ 50 | too few trades to say anything |
| WARN | Cost model source | fallback published schedule | = coinbase api | fees assumed from the published schedule ($0-$10K: 40/60 bps); 1 cost model(s) in this family are not live reads. Configure COINBASE_API_KEY and COINBASE_API_SECRET and re-run cost_model_refresh |
| PASS | Venue consistency | coinbase advanced | = coinbase advanced published | bars from coinbase_advanced, fees from coinbase_advanced_published |
| WARN | Breadth · effective tests | 2.94 | ≥ 3 | 22 assets provide 2.9 effective tests (mean pairwise correlation 0.5267), M_eff by unlabelled; the estimator was not recorded, so this is judged against the participation_ratio hurdle — entropy-based M_eff on the same universe measures ~2.3x higher and is not comparable |
The referee's answer is no. What failed: deflated sharpe; probability of backtest overfitting; t-statistic; backtest length; regime survival; trades. A strategy fails when the evidence cannot distinguish it from the best of 2 attempts at nothing — not because it lost money. 2 warnings reported but not blocking.
What this page does not show: the strategy's parameters. The owner published the verdict, not the spec — only its hash prefix 66a9f9acc391, which is what the ledger binds this result to. Spec still open; holdout registered.
Same seven blocking checks, on your strategy, from any MCP client. The ledger counts every trial you run; the holdout opens once.