# Verification checks (P4)
| # | check | outcome |
|---|---|---|
| 1 | Clean re-run (deleted results.json, figure1.png, exclusions.csv) regenerates byte-identical results.json | PASS (cmp identical) |
| 2 | Re-run after exclusion-ledger refactor gives identical results.json | PASS (diff empty) |
| 3 | Exclusion ledger: 319 − 5 − 73 − 1 = 240 = n_analysed; each n_before = previous n_after | PASS |
| 4 | Every quarter 1959Q4–2019Q4 has 3 non-missing UNRATE months; 241 GDP levels; T = 240 | PASS (asserts in analysis.py) |
| 5 | first_quarter = 1960Q1, last_quarter = 2019Q4 (no data beyond window used in regression) | PASS |
| 6 | Manual Bartlett NW sandwich equals statsmodels HAC SE to 1e-10 | PASS (assert) |
| 7 | se_iid > 0, se_hac > 0 | PASS (0.01996, 0.03583) |
| 8 | CI [−0.3260, −0.1855] contains beta_hat −0.2558; symmetric: (lo+hi)/2 = beta_hat | PASS |
| 9 | se_hac > se_iid consistent with positive residual lag-1 autocorrelation (0.389) | PASS (plausible direction) |
| 10 | okun_negative_5pct = (beta<0 and p_hac<0.05): beta<0, p≈9.5e-13 → true | PASS |
| 11 | Magnitudes: mean_growth 0.754 %/q (~3.0 % annualised), alpha 0.184 pp, implied growth for stable u = 0.72 %/q — plausible for U.S. 1960–2019 | PASS |
| 12 | Bandwidth sensitivity: se_hac in [0.0288, 0.0371] for L ∈ {0,…,12}; inference unchanged | PASS |
| 13 | Input checksums match README | PASS |