# Checks (P4)

1. Clean re-run reproducibility: ran analysis.py twice, results.json byte-identical (cmp) — PASS.
2. Counts: n_input_rows 2498; table1 n_days sum = 2498; exclusion ledger 2498 − 1 = 2497 changes used for baseline_metric — PASS.
3. n_treated = 0 confirmed independently in the audit (value_counts) — PASS. That's why effect, CI, model_metric and n_analysed are null.
4. Chronology: sample_start 2010-01-04 < sample_end 2019-12-31 — PASS. post_cut agrees with the date ≥ 2020-03-04 rule on 0 mismatching rows — PASS.
5. Magnitude: baseline_metric −0.043 bp/day ≈ (1.68 − 2.75)·100/2497 = −0.0429 — PASS.
6. CI-contains-estimate: not applicable (all null) — PASS (allowed when infeasible).
7. Gate runs: recorded in phases.json.