1# Verification checks21. Clean re-run (results.json and exclusions.csv deleted, script re-run): byte-identical results.json — PASS.32. Manual CR1 cluster variance equals statsmodels se_cluster to 1e-10 (assert in script) — PASS.43. se_ols > 0, se_cluster > 0 — PASS.54. beta_ci_low < beta_hat < beta_ci_high, CI symmetric about beta_hat — PASS.65. n_input_rows (2000) - removed (0) = n_analysed (2000); 50 clusters x 40 = 2000 — PASS.76. Magnitude vs baseline: se_cluster/se_ols = 3.03, plausible with firm-level shocks and within-firm correlated x; beta_hat identical under both (only SE differs) — PASS.8