Glossary
Data validation
Checking that data meets explicit, predefined rules for type, range, format, or consistency before it is used or trusted.
Data validation is the process of checking that data conforms to explicit, predefined rules before it is loaded, transformed, or trusted, for example that an email field matches a valid pattern, an order_date is not in the future, or a foreign key actually exists in the referenced table. Failing records are typically rejected, flagged, or routed to a quarantine table rather than allowed downstream silently.
Validation differs from data profiling, which describes what a dataset actually contains without judging it; validation applies rules and produces a pass or fail result. It is usually implemented as automated checks inside a data pipeline, run either at ingestion, after transformation, or both, and is one of the concrete mechanisms behind broader data quality programs.
Validation matters because bad data caught late is far more expensive than bad data caught early: a malformed date that reaches a finance report can trigger a restatement, while the same error caught at ingestion is a one-line fix. Rules also need to evolve alongside the data; a common pitfall is writing validation rules once and letting them go stale as schema evolution changes what "valid" means, so checks either miss new problems or reject legitimate new data. Continuous data observability helps catch the gap between the two.
Last reviewed September 22, 2026