| 1 |
Clean re-run of out/analysis.py (deleted outputs) regenerates byte-identical results.json |
PASS |
| 2 |
Exclusion ledger: 5217 − 215 − 1 = 5001 = n_analysed = n_train + n_test (2500 + 2501) = sum of table1 n |
PASS |
| 3 |
ANOVA F cross-checked with statsmodels OLS dy~C(weekday) overall F (0.26347, p 0.9015, df 4/4996) |
PASS (identical to 1e-12) |
| 4 |
Monday contrast Welch CI contains estimate; CI agrees with HC1 regression CI and Newey-West(5) CI to ~0.01 bp |
PASS |
| 5 |
SEs positive (table1 se_bp, effect_se_welch) |
PASS |
| 6 |
Chronology: max train date 2009-12-31 < min test date 2010-01-04; weekday means computed from train only (train_weekday_means_bp) |
PASS |
| 7 |
Magnitudes: baseline RMSE 4.88 bp vs full-sample sd of changes 5.72 bp (2000-09 more volatile) — plausible; model/baseline ratio ≈1.002 |
PASS |
| 8 |
Conclusion rule: ANOVA p=0.90 ≥ 0.05 → no_detectable_effect; Monday CI also covers 0 (consistent) |
PASS |
| 9 |
Equal-variance assumption of classical ANOVA: Levene p≈1e-7 rejects equal variances → ANOVA kept as prescribed; Kruskal-Wallis (p 0.47) and HC/HAC contrast CIs agree in not rejecting |
NOTED (limitation) |