# Verification checks (P4)
1. Clean rerun: deleted results.json/figure1.png/table1.csv/exclusions.csv, re-ran analysis.py; results.json byte-identical (cmp) — PASS.
2. Input SHA-256 equals README value — PASS.
3. Baseline identity: mean one-day change = 100·(y_last − y_first)/(n−1) = -0.0428514 bp, equals baseline_metric — PASS.
4. table1 n_days sum = 2498 = n_input_rows — PASS.
5. Exclusion ledger: 2498 − 2498 removed = 0 rows in effect estimate; n_analysed = null (no observations) — consistent — PASS.
6. sample_start (2010-01-04) <= sample_end (2019-12-31) — PASS.
7. n_treated = 0 equals count of post_cut==1 — PASS; hence effect/CI/model_metric null, conclusion premise_invalid as TASK.md requires — PASS.
8. CI contains estimate / positive SE: not applicable (all null) — N/A.
9. Gate runs: recorded in phases.json.