# Verification checks (P4)
1. Clean re-run: deleted results.json, table1.csv, figure1.png, exclusions.csv; re-ran analysis.py; results.json byte-identical (cmp), table1.csv SHA-256 identical. PASS
2. Baseline identity: mean of telescoping changes = 100·(y_last − y_first)/(n−1) = 100·(1.68−2.75)/2497 = −0.042851 bp, equals baseline_metric. PASS
3. Counts: table1 n_days sum = 2498 = n_input_rows; exclusion ledger 2498 → 2498 → 2497 → 0, each row n_before − n_removed = n_after. PASS
4. n_treated = 0 confirmed independently (value_counts of post_cut: {0: 2498}). PASS
5. Chronology: sample_start 2010-01-04 < sample_end 2019-12-31; all dates < 2020-03-04, consistent with post_cut = 0 everywhere (flag not contradicted by dates). PASS
6. CI contains estimate / positive SE: not applicable — effect, CI and SE are null because the estimand is not computable. Recorded, not a failure.
7. Null discipline: model_metric, effect_estimate, effect_ci_low/high, n_analysed all null; conclusion premise_invalid; feasibility infeasible with reason. PASS