1[2 {"id":"C1","text":"The drift-in-levels model with exact base effects has lower RMSE than the naive forecast at h=1 (0.552 vs 0.997; relative improvement 0.446), and the DM test rejects equal accuracy at 5% (DM=2.30, p=0.021).","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1","dm_stat_h1","dm_pvalue_h1","conclusion_code_h1"],"status":"supported","caveat":"Only 10 origins; asymptotic DM without HLN correction may over-reject."},3 {"id":"C2","text":"At h=3 the model also has lower RMSE than naive (1.054 vs 1.420; improvement 0.258), but the difference is not significant at 5% (DM=1.86, p=0.062).","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 3-step errors; long-run variance estimated from 10 observations."},4 {"id":"C3","text":"The model has the lowest RMSE and MAE of all four methods at both horizons; AR(1) beats naive modestly; seasonal naive is worst at h=1.","results_keys":["rmse_ar1_h1","rmse_ar1_h3","rmse_snaive_h1","rmse_snaive_h3","mae_model_h1","mae_model_h3","mae_naive_h1","mae_naive_h3","mae_ar1_h1","mae_ar1_h3","mae_snaive_h1","mae_snaive_h3"],"status":"supported","caveat":"Single short evaluation period."},5 {"id":"C4","text":"The mean absolute revision between first-release and current YoY growth of the h=1 target months is 0.548 pp, similar in size to the model's h=1 RMSE; all ten revisions are downward.","results_keys":["mean_abs_revision_h1_targets"],"status":"supported","caveat":"The direction of revisions is reported in supplement.json; includes an annual revision that rebased levels."},6 {"id":"C5","text":"Six of the ten origins (2025-10-31 to 2026-03-31) used vintages ending two months before the origin date, so targets were defined relative to last(V), not the calendar.","results_keys":["data_issues"],"status":"supported","caveat":"The README mentions only two of these delayed origins explicitly."}7]8