1[2 {"id":"C1","text":"Over 35 evaluated origins at h=1, AR2_diff had higher RMSE (0.1199) than the naive forecast (0.1171); relative improvement -0.0235.","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1","n_evaluated_h1"],"status":"supported","caveat":"Final-vintage data; short window."},3 {"id":"C2","text":"At h=3, AR2_diff had slightly higher RMSE (0.1598) than naive (0.1586); relative improvement -0.0076.","results_keys":["rmse_model_h3","rmse_naive_h3","rel_improvement_rmse_h3","n_evaluated_h3"],"status":"supported","caveat":"Overlapping h=3 errors; 35 pairs."},4 {"id":"C3","text":"MAE of AR2_diff also exceeded naive MAE at both horizons.","results_keys":["mae_model_h1","mae_naive_h1","mae_model_h3","mae_naive_h3"],"status":"supported","caveat":"Differences are a few thousandths of a percentage point."},5 {"id":"C4","text":"The fitted AR coefficients are near zero, making the model nearly a random walk.","results_keys":["ar_coefficients_last_window"],"status":"supported","caveat":"Attribution to COVID outliers is a hypothesis, not tested."},6 {"id":"C5","text":"The DM test indicates higher model loss at h=1 (p=0.017) but not at h=3 (p=0.18); p-values are only indicative because the models are nested and the sample is small.","results_keys":["dm_stat_h1","dm_pvalue_two_sided_h1","dm_stat_h3","dm_pvalue_two_sided_h3"],"status":"supported","caveat":"Standard DM test is mis-sized for nested models; Clark-West not run."},7 {"id":"C6","text":"Two (origin,horizon) pairs targeting the missing 2025-10 month are excluded, leaving 35 evaluated pairs per horizon; 941 observations in the last training window.","results_keys":["n_evaluated_h1","n_evaluated_h3","n_analysed","n_input_rows","data_issues"],"status":"supported","caveat":""}8]9