# Verification checks (P4)
1. Clean re-run: deleted results.json/analytic.csv/exclusions.csv, re-ran analysis.py; SHA-256 of all three identical — PASS.
2. analytic.csv rows (1800) == n_analysed; distinct firms (300) == n_firms — PASS.
3. analytic.csv contains no rating_2025 column — PASS.
4. Exclusion ledger: n_before − n_removed = n_after on every row; chained; first n_before = 2140 = n_input_rows; last n_after = 1800 = n_analysed — PASS.
5. se_leverage_cluster > 0; 95% cluster CI [0.184, 0.344] contains beta_leverage 0.264 — PASS.
6. default_rate equals mean of default_next in analytic.csv (0.1211; 218/1800) — PASS.
7. Chronology: max origin year in sample 2021 (outcome year 2022 ≤ data cut); no duplicated firm-year keys; no missing values — PASS.
8. Magnitude vs baseline: beta 0.264 per unit leverage; 1 SD of leverage (~0.21) ⇒ ~5.5 pp vs base rate 12.1% — plausible. Cluster SE (0.0408) > iid SE (0.0358), as expected with within-firm correlation — PASS.