When you run a data subset operation, the task selects the records in an object based on the filter criteria. The filtered records of the object can contain child records. The child records might have references to other records from the same parent object. In such cases, the task loads additional records along with filtered record to maintain referential integrity.