1[2 {"id":"C1","text":"The employment-augmented direct ARDL model has lower RMSE and MAE than the naive forecast at h=1 (RMSE 0.743 vs 0.814, relative improvement 0.087).","results_keys":["rmse_model_h1","rmse_naive_h1","mae_model_h1","mae_naive_h1","rel_improvement_rmse_h1"],"status":"supported","caveat":"DM p=0.048, borderline; current-vintage data, timing simulation only."},3 {"id":"C2","text":"At h=3 the model has lower RMSE and MAE than naive (RMSE 0.902 vs 1.140, relative improvement 0.208).","results_keys":["rmse_model_h3","rmse_naive_h3","mae_model_h3","mae_naive_h3","rel_improvement_rmse_h3"],"status":"supported","caveat":"Not statistically significant (DM p=0.18)."},4 {"id":"C3","text":"The h=1 improvement is borderline significant and the h=3 improvement is not significant under a Diebold-Mariano test.","results_keys":["dm_stat_h1","dm_pvalue_h1","dm_stat_h3","dm_pvalue_h3"],"status":"supported","caveat":"48 origins, overlapping h=3 errors."},5 {"id":"C4","text":"The size of the gain is sensitive to excluding the COVID period from training (0.012 at h=1 and 0.304 at h=3 without exclusion); only its sign holds in both specifications.","results_keys":["sens_rel_improvement_rmse_nocovidexcl_h1","sens_rel_improvement_rmse_nocovidexcl_h3"],"status":"supported","caveat":"Only two specifications compared."},6 {"id":"C5","text":"All forecasts respect the publication-lag rules (GDP <= t-2, employment <= t-1); at the last origin these are 2026-02-01 and 2026-03-01.","results_keys":["last_target_date_available","last_predictor_date_available","last_origin"],"status":"supported","caveat":"Verified via forecasts.csv date columns and in-script asserts."}7]8