# Verification checks (P4)

1. Clean re-run: deleted out/results.json, ran analysis.py twice; byte-identical outputs (cmp) — PASS.
2. Counts: 5217 − 215 − 1 = 5001 = n_analysed = n_train + n_test (2500+2501) = sum of table1 n — PASS. Exclusion ledger sums — PASS.
3. CI contains estimate (−0.319 < 0.050 < 0.420) — PASS. All SEs > 0 — PASS.
4. Monday contrast recomputed from table1 weighted means: 0.050170 = effect_estimate — PASS.
5. Independent cross-check with statsmodels OLS dy~C(weekday): F=0.26347, p=0.90151, identical to scipy f_oneway — PASS. HC1 regression on Monday dummy: 0.0502, CI [−0.319, 0.419], agrees with Welch — PASS.
6. Chronology: max train date < min test date (assert in code); weekday means estimated from train only — PASS.
7. Magnitudes: RMSE ≈ 4.88 bp is plausible for daily 10y changes; model RMSE (4.8838) > baseline (4.8757), ratio 1.0017 — consistent with no effect.
8. Conclusion rule: ANOVA p=0.90 ≥ 0.05 → no_detectable_effect — PASS.