spb/anomaly-atlas Public License

1---2project: anomaly-atlas3document: expH_oos_stability/analysis4author: Simon-Pierre Boucher5contact: contact@spboucher.ai6data_source: hfmarketdata.io7created: 2026-08-128modified: 2026-08-129status: reviewed10---1112# Analysis — expH_oos_stability1314Run: `results/expH_oos_stability/20260812T072518Z/results.json`. The15validation split (2016-2021), frozen before the first scan, was opened here16for the first time (418 requests, 18.8M rows). The holdout (2022→) remains17sealed. All four pre-registered questions resolved:1819| Question | Train | Validation | Verdict |20|---|---|---|---|21| Q-b nothing survives κ=1 | 0 survivors | **0 survivors** | negative finding **replicates** |22| Q-a CKX 1day net@0.25 | κ*=0.28 | +16.3 bp/d CI [1.4, 26.5], but 2019-21 CI [−19.5, +22.0] | **fails robustness clause** — not confirmed |23| Q-c ES→SPY fresh corr | −0.032 (FDR, splice-invariant) | **+0.003** (p=0.68) | **does not replicate** |24| Q-d daily reversal family | VR30 excess −0.143 | **+0.023** | crisis-regime story **confirmed** |2526## Reading27281. **The ladder closes on the honest outcome.** Artifact nulls → search29   correction → costs → OOS: nothing survives all four. This is the30   charter's anticipated result ("a well-documented 'it's all artifacts'31   is a success"), now demonstrated rather than assumed — and it becomes32   the atlas's first Level-2 entry.332. **CKX was exactly what it looked like.** The lone cost survivor decays34   within validation (κ* 0.445 → 0.263), fails its pre-registered35   two-sub-period clause, and lives in a name whose capacity rounds to36   zero. Not escalated; the sealed holdout is not spent on it.373. **ES→SPY is a lesson in regime-fragility**: a splice-invariant,38   FDR-significant train effect that flips sign OOS. Splice-invariance39   guards against one artifact — it says nothing about stationarity.404. **The artifact replicates even though the anomalies do not.** SPY still41   leads SPX on validation (+0.080 vs +0.132 in train) — content staleness42   is structural. That contrast (artifacts stable, "alphas" not) is the43   project's cleanest single sentence.4445## Hand-off4647Two atlas entries created via `tools/new_finding.py` (F001 negative48finding, Level 2; F002 SPX artifact demonstration, Level 2). Candidate 0249(turn-of-month) and candidate 01 (intraday momentum) keep their protocols50for future work; the holdout stays sealed until a claim earns it.51