# Verification checks (P4)

1. Clean re-run: removed results.json, figure1.png, table1.csv and exclusions.csv, re-ran analysis.py, and cmp against the earlier results.json → IDENTICAL. PASS
2. Input hash unchanged after the run (2cdda6cc...). PASS
3. table1 n_days sum = 2498 = n_input_rows. PASS
4. Exclusion ledger: 2498 − (1 + 2497) = 0 analysed rows, matching n_analysed = null (nothing analysed). PASS
5. baseline_metric = 100*(y_last − y_first)/2497 = −0.042851 bp (telescoping identity). PASS
6. n_treated = 0 matches the audit; conclusion premise_invalid ⇔ feasibility infeasible, and every inestimable key is null. PASS
7. Code-path smoke test (temp copy in /tmp with an artificial flag, results thrown away and NOT reported): the effect, Welch CI, RMSE and n_analysed all compute; the CI contains the estimate, se > 0, and n_analysed = 120 + n_post. PASS (this checks only that the code runs; the numbers have no meaning)
8. Magnitudes: baseline mean change (~0 bp) is small next to the daily sd of 4.24 bp, which is plausible. PASS