Quarantine sets aside the rows that fail the quality checks, in a separate table or file with the reason for rejection, rather than deleting them or loading them. The pipeline continues with the valid data, and the rejects can be analyzed, then corrected or replayed.