# Verification checks (P4)

• Clean re-run (outputs deleted, analysis.py re-run): results.json and forecasts.csv byte-identical: PASS
• Leakage test: all values after 2024-12-01 set to 9.9 in a temp copy; forecasts for origins <= 2024-12-01 identical, later ones change: PASS
• 144 ledger rows: PASS
• unique (origin,h,model): PASS
• target=origin+h months: PASS
• actuals match data: PASS
• n_eval = non-missing actuals: PASS
• naive = last obs <= origin: PASS
• rmse/mae naive h1 recomputed from ledger: PASS
• rmse/mae model h1 recomputed from ledger: PASS
• rel_improvement h1 identity: PASS
• rmse/mae naive h3 recomputed from ledger: PASS
• rmse/mae model h3 recomputed from ledger: PASS
• rel_improvement h3 identity: PASS
• RMSE>=MAE>0: PASS
• exclusion ledger: 945-1-3=941=n_analysed: PASS
• first_origin<last_origin; last target 2026-09 within data: PASS
• naive RMSE h3 > h1 (plausible magnitude): PASS
• Descriptive: UNRATE min 3.5 / max 4.5 over 2023-07..2026-09 (non-missing): computed, cited in report Uncertainty.