step,rule,n_before,n_after,n_removed,rationale
1,candidate forecast-actual pairs = 10 origins x 2 horizons x 4 models,80,80,0,all targets last(V)+h have a first-release actual; nothing excluded
2,pairs for own model (evaluated for RMSE/MAE/DM),80,20,60,baseline rows (60) are evaluated separately in table1; n_analysed counts model pairs only