step,rule,n_before,n_after,n_removed,rationale
1,first_release.csv: keep only rows that are targets last(V)+h of some origin,28,13,15,rows outside the evaluation design are not needed as actuals (not an error-driven exclusion)
2,forecast-actual pairs for own model: drop pairs with missing forecast or actual,20,20,0,no pair is missing; nothing removed