# Verification checks (P4)
- Clean re-run (deleted results.json, diagnostics.json, exclusions.csv; re-ran analysis.py): results.json byte-identical — PASS
- bias_analytical == -(1+3*0.8)/30: PASS
- bias_mc == rho_hat_mean - 0.8: PASS
- bias_mc_se == rho_hat_sd_mc/sqrt(n_analysed): PASS
- SEs positive: PASS
- discrepancy_se_units recomputes from components: PASS
- 95% MC CI contains bias_mc: PASS
- se_asymptotic == sqrt(0.36/30): PASS
- n_analysed == n_replications - excluded (20000 = 20000 - 0): PASS
- Magnitude: bias_mc negative and within factor 2 of analytical (ratio ≈ 1.09): PASS
- Seed equals design seed: PASS
No failed checks.