1# Verification checks (2026-10-30)23| # | check | outcome |4|---|---|---|5| 1 | Input SHA-256 match README/bundle.json | PASS |6| 2 | Quarterly series recomputed from monthly file (max abs dev < 1e-5; asserted in script) | PASS (≤5e-7) |7| 3 | Clean re-run (deleted results.json, forecasts.csv) gives byte-identical files | PASS |8| 4 | forecasts.csv has 240 rows, 40 per (model, horizon) | PASS |9| 5 | target_date = origin_date + h quarters for all rows | PASS |10| 6 | actual equals infl_annualised of target quarter | PASS |11| 7 | snaive forecast equals value of quarter t+h−4 (≤ origin) | PASS |12| 8 | Look-ahead test: adding +5 to all inflation values after 2020Q1 leaves model forecasts at origins ≤ 2020Q1 unchanged and changes later ones | PASS |13| 9 | RMSE ≥ MAE for every model/horizon | PASS |14| 10 | loss_diff_se_h4 > 0; dm_stat = mean/se | PASS |15| 11 | d̄ recomputed independently from forecasts.csv (2.81795) equals loss_diff_mean_h4 | PASS |16| 12 | hac_lags_h4 = 4 ≥ h−1 = 3 | PASS |17| 13 | n_raw (127) − removed (1) = 126 usable quarters; last estimation sample = 122 obs to 2025Q3 − p(=1) lags = 121 = n_analysed | PASS |18| 14 | Magnitudes vs baseline: model RMSEs (2.19–2.46) below snaive (2.85–2.88), same order of magnitude as series sd (2.39) | PASS (plausible) |1920Note: sample autocorrelations of d_t at lags 1–4: 0.21, −0.05, −0.29, −0.41. Negative autocorrelation at lags 3–4 makes the SE sensitive to the lag/kernel choice (see sensitivity_h4).21