Two datasets disagree about active customers. How would you reconcile them?

Agree on a precise definition of active customer, including date window, timezone, account status, and treatment of trials or deleted accounts. Compare source lineage, refresh times, identity mappings, and join cardinalities. Reconcile counts through progressively narrower stages and inspect specific records present in only one result. Watch for duplicate customer identifiers and event-based joins that multiply rows. Document the root cause, create a reproducible reconciliation query, and assign an owner for the metric. Choosing whichever count looks more plausible does not resolve the inconsistency.