Asset Check

An asset check evaluates a named property of a data asset, such as a required column or an allowed value range. The result is evidence about that property and the input version actually examined.

Check execution and downstream gating must be configured. In Dagster, a check is not blocking by default; blocking=True with an error-level failure can prevent downstream materialization when the check is included in the execution path.

A check can run after its upstream asset has already been materialized. Blocking downstream work does not undo that output or revoke access. Stage and validate before publication if invalid candidates must stay invisible.

A passing nonnegative-value test does not detect every missing row or incorrect business definition. Define independent completeness and freshness evidence, and verify that skipping or separately selecting checks cannot bypass the intended publication rule.

Suppose an asset contains only nonnegative sales but omits an entire store. The range check passes while completeness fails. Record each check’s input version, measured value, threshold, and result; use an independent expected-store list for the missing-store check. Decide whether a warning is acceptable or publication must stop, and test that decision with a deliberately failing candidate.

See Asset-Oriented Orchestration: Dagster and Prefect for worked examples.

Further reading

• Dagster: Testing assets with asset checks