Duplicated alternative-data records

Near-identical vendor records occur in both training and validation under different IDs. How would you diagnose and prevent contamination?

Compare timestamps, entity keys and content fingerprints, then inspect near-duplicate clusters. Define deduplication and grouping rules using features available without validation labels. Keep related records in one permitted split or remove redundant observations before evaluation under a documented policy. Confirm that repeated legitimate events are not discarded simply because they look similar. Distinct database IDs do not imply independent observations.