1[2 {"id":"C1","text":"Over 60 origins (2021-09 to 2026-08), the ridge forecast has a 36.5% lower RMSE than the naive forecast.","results_keys":["rmse_ridge","rmse_naive","rel_improvement_vs_naive","n_origins","first_origin","last_origin"],"status":"supported","caveat":"Current-vintage data; publication lags ignored."},3 {"id":"C2","text":"The ridge forecast does not beat the expanding-mean forecast: its RMSE is 3.1% higher and its MAE is also higher.","results_keys":["rmse_ridge","rmse_mean","mae_ridge","mae_mean","rel_improvement_vs_mean"],"status":"supported","caveat":"Short 60-month evaluation window."},4 {"id":"C3","text":"The ridge model's gain over the naive forecast is statistically significant (DM p=0.0014), while its difference from the expanding mean is not (DM p=0.59).","results_keys":["dm_stat_vs_naive","dm_pvalue_vs_naive","dm_stat_vs_mean","dm_pvalue_vs_mean"],"status":"supported","caveat":"Diebold-Mariano test with HLN correction; small sample; the models are nested-like, so the DM test can be conservative."},5 {"id":"C4","text":"Standardisation statistics and the target mean were computed from each origin's training window only (expanding), using 413 training rows at the last origin.","results_keys":["normalisation_window","n_analysed","lambda","n_features"],"status":"supported","caveat":"Verified by a leakage perturbation test (out/state/checks.md)."}6]7