equity_dedup_reweight on aggressive KILL
Did not survive the pre-registered gate. This is a first-class result, not an error: the honest thing to do with a hypothesis that fails is publish the kill.
Equity-cluster de-duplication, expression B (reweight). The aggressive book's QQQ/SPY/MTUM sleeves run 0.77 to 0.93 realized correlation, so three of its nine sleeves are close to one bet. This expression keeps all three instruments and divides each one's declared risk-budget tilt by the cluster size (3), so the correlated cluster collectively draws one sleeve's budget rather than three and the freed budget flows to the book's genuine diversifiers through the same inverse-vol pipeline. Unlike expression A it drops no instrument, so it carries no which-instrument-won confound. Tests whether raising the number of independent bets, with no new instrument and no new data, clears the full deflation / walk-forward / PBO / incremental-DSR gate.
The pre-registered gate
| Deflated incremental edge | -0.089 FAIL (threshold +0.000) |
| Walk-forward out-of-sample uplift | +0.000 FAIL (threshold +0.000) |
| Probability of backtest overfitting (PBO) | 0.41 PASS (threshold 0.50) |
No-lookahead probe: PASSED (divergence date 2024-03-27)
The underlying comparison
In-sample figures are the best-of-menu figure on one price history and are upward-biased by selection. Where they disagree with the deflated or walk-forward out-of-sample figures above, those win.
| In-sample Sharpe, candidate vs baseline | +0.949 vs +1.038 (uplift -0.089) |
| Walk-forward out-of-sample Sharpe, candidate vs baseline | +1.000 vs +1.000 |
| PBO combinations evaluated | 924 |
| Correlation to the plain baseline book | +0.896 (n=3851) |
Trust ladder
Current rung: UNVERIFIED. A hypothesis reaches a stakeholder surface as a claim only once it clears VERIFIED; see the verification page for what each rung means.
| validated | FAIL research loop verdict: KILL |
| lookahead_probe | PASS probe passed (divergence 2024-03-27) |
| panel_pinned | PASS panel_manifest_hash=f75ad5dbd3528e36 |
| harness_calibrated | FAIL no calibration report supplied for this profile |
| adversarial_red_team | FAIL no red-team artifact for this candidate: the 'How it dies' fan (report 1.2.8) auto-runs on SHIP-OFF and is required for VERIFIED (`ballast.verification.redteam`) |
Check the source
Every number above is read straight from these committed files, nothing recomputed for this page:
data/research/equity_dedup_reweight-aggressive.json # the ORIGINAL pre-registered verdict, machine readable (never rewritten) data/research/equity_dedup_reweight-aggressive.md # the same original verdict, written out data/research/equity_dedup_reweight-aggressive.verification.json # trust ladder rung for the original verdict