1# Verification checks (P4)23- Clean re-run (deleted results.json/forecasts.csv, ran twice): byte-identical outputs — PASS4- ledger has 144 rows: PASS5- 36 unique origins, 2023-07-01..2026-06-01: PASS6- target = origin + h months: PASS7- naive = last non-missing obs <= origin: PASS8- actuals match data: PASS9- n_evaluated = 36 - missing targets: PASS10- RMSE >= MAE > 0: PASS11- rel_improvement recomputed: PASS12- exclusion ledger sums: 945 - removed = n_analysed: PASS13- h3 error >= h1 error (plausible magnitudes): PASS14- model forecasts within 0.2pp of naive (consistent with near-zero AR coefs): PASS15- first_origin < last_origin: PASS16- Information rule: each origin's estimation uses `s.loc[:t]` only (code inspection); targets strictly after origins (checked above) — PASS17