spb/anomaly-atlas Public License

1{2  "anomaly_id": "F001-short-horizon-survival",3  "title": "Nothing in the searched universe survives the full validation ladder",4  "summary": "Across 22 pre-registered hypotheses (372 searched rules: reversion, lead-lag, calendar; 2000-2016 train), every effect that beats the measured artifact nulls and Hansen-SPA search correction dies at a fraction of its own half-spread (median breakeven = 1.1% of one half-spread; zero intraday rules survive paying the full half-spread), and this negative result REPLICATES on the untouched 2016-2021 validation split (zero net survivors at full spread; the lone marginal 0.25x survivor fails its pre-registered sub-period clause; the splice-invariant ES-SPY train effect flips sign OOS; the daily 2008-2015 reversal family was crisis-regime).",5  "negative_result": true,6  "key_numbers": {7    "searched_rules": 372,8    "naive_significant": 232,9    "fdr_survivors": 226,10    "spa_step1_survivors": 68,11    "median_breakeven_kappa": 0.0114,12    "net_survivors_at_full_halfspread_train": 0,13    "net_survivors_at_full_halfspread_validation": 014  },15  "evidence": [16    "results/expC_reversion_scan/",17    "results/expD_leadlag_scan/",18    "results/expE_calendar_scan/",19    "results/expF_multiple_testing/",20    "results/expG_cost_frontier/",21    "results/expH_oos_stability/"22  ],23  "reproduce": "make setup && python experiments/micro/expF_multiple_testing/benchmark.py (cache-served); full chain expA->expH per experiments/micro/*/hypothesis.md",24  "author": "Simon-Pierre Boucher",25  "contact": "contact@spboucher.ai",26  "data_source": "hfmarketdata.io",27  "confidence_level": 228}29