step,rule,n_before,n_after,n_removed,rationale
1,drop rows with missing UNRATE from the observation set (2025-10-01),945,944,1,empty value field; month kept in forecast ledger as target with empty actual; interpolated only as a model lag input at origins >= 2025-11
2,keep only observations dated <= last origin 2026-06-01 (last training window),944,941,3,2026-07..2026-09 are evaluation targets only and never enter any training window