Disable tc2-league-filter — drift analysis shows no alpha
One-variable change: flip leagueExclude from ["segunda", "la-liga", "ligue-2"] to [] in DEFAULT_EVAL_CONFIG at lib/backtest/bet-evaluator.ts:294.
One-variable change: flip leagueExclude from ["segunda", "la-liga", "ligue-2"] to [] in DEFAULT_EVAL_CONFIG at lib/backtest/bet-evaluator.ts:294.
Signal was accepted 2026-03-28 on Entry-adj marginal ROI +1.2% (with manual overrides on Gate 5 bootstrap significance p=0.1567 and Gate 8 suspicious-N). The 2026-04-14 drift sweep showed marginal dropped to +0.04%, prompting a per-league ROI analysis.
Per-league findings on all-time data (filter OFF): - Excluded aggregate: N=2,719, entry-adj ROI +6.58%, CLV +10.99% - Rest of stack: N=16,984, entry-adj ROI +6.90%, CLV +11.80% - Gap: 0.32pp (within bootstrap noise)
Per-league current season (2025): - Segunda: N=146, entry +6.25% (fine) - La Liga: N=92, entry -8.77% (losing) - Ligue 2: N=61, entry +16.37% (actively winning)
The filter is costing us alpha on Segunda and Ligue 2 while only catching La Liga's current-season weakness, which is noise-dominated at N=92 and would be better handled by a narrower signal if real. Disabling entirely restores ~2,700 bets of exposure to leagues that are not structurally unprofitable.
Registry entry flipped to status="drift-flagged" with full statusHistory. Analysis script preserved at scripts/analyze-tc2-league-filter-drift.ts for reproducibility.
Full context: ~/.claude/plans/2026-04-15-drift-findings.md ~/.claude/plans/2026-04-14-drift-sweep-log.txt
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>