# Verification checks (P4)
Re-run from clean state (deleted results.json, table1.csv, figure1.png, exclusions.csv): results.json byte-identical (cmp) — PASS
- naive CI contains estimate : PASS
- effect CI contains estimate : PASS
- positive effect_se, baseline_metric, model_metric : PASS
- exclusions sum: n_raw - removed = n_analysed : PASS
- n_input_rows matches wc : PASS
- model RMSE <= baseline RMSE (nested model) : PASS
- conclusion consistent with CI : PASS
- robust vs conventional SE similar (ratio) : PASS
- omitted-variable identity: naive = effect + skill_coef*cov(T,S)/var(T) (-0.131698 vs -0.131698) : PASS
- sensitivity estimates within main CI : PASS
All checks passed; none required fixing.