1[2 {"id":"C1","text":"At h=1 the base-effect random-walk-with-drift model has lower RMSE than the naive forecast (0.550 vs 0.997 pp; 44.8% improvement), and the DM test as specified rejects equal accuracy at 5% (DM=2.43, p=0.015).","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1","dm_stat_h1","dm_pvalue_h1","conclusion_code_h1"],"status":"supported","caveat":"n=10; one origin contributes ~45% of the loss differential; HLN-corrected p=0.047 (sensitivity, aux file)."},3 {"id":"C2","text":"At h=3 the model has lower RMSE than naive (1.142 vs 1.420 pp; 19.6% improvement) but the difference is not significant (DM=1.55, p=0.12).","results_keys":["rmse_model_h3","rmse_naive_h3","rel_improvement_rmse_h3","dm_stat_h3","dm_pvalue_h3","conclusion_code_h3"],"status":"supported","caveat":"Overlapping multi-step errors; long-run variance from 10 points is imprecise."},4 {"id":"C3","text":"Among baselines, AR(1) is slightly better than naive at both horizons and seasonal naive is worst at h=1 but beats naive at h=3.","results_keys":["rmse_ar1_h1","rmse_ar1_h3","rmse_naive_h1","rmse_naive_h3","rmse_snaive_h1","rmse_snaive_h3"],"status":"supported","caveat":"No significance tests among baselines."},5 {"id":"C4","text":"First-release YoY growth of the h=1 target months differs from the current-vintage growth by 0.548 pp on average, about the size of the model's h=1 RMSE.","results_keys":["mean_abs_revision_h1_targets","rmse_model_h1"],"status":"supported","caveat":"Current vintage includes a large level revision; rankings against current data not evaluated."},6 {"id":"C5","text":"All forecasts are strictly real-time: each uses only the RSAFS vintage dated at its origin, and no target month is present in that vintage.","results_keys":["n_analysed","first_origin","last_origin"],"status":"supported","caveat":"Gate check origin_date<target_date fails for h=1 rows by construction of the TASK target definition (publication lags), documented in report."}7]8