Quality Dimension Threshold
The stated numeric level at which a dataset is fit for its purpose on a given quality dimension, plus what happens when it is not met.
Completeness, accuracy, timeliness, consistency, validity and uniqueness become useful only when each carries a number and a consequence. "Data quality is important" is not a control; "customer email is populated on at least 99.5% of active records, checked daily, and the pipeline fails below 98%" is.
Two thresholds rather than one is the arrangement that works. A warning level notifies the owner and a failure level stops downstream processing. The gap between them is what prevents a pipeline halting over a marginal fluctuation while still catching a genuine collapse.
Setting the numbers has to be a fitness-for-purpose conversation with the consumer, not a target chosen by the data team. Ninety-five percent completeness on a marketing attribute is fine; the same figure on a field feeding a regulatory return is not.
The decision that gets deferred and should not is what happens on failure. Halting the pipeline protects downstream correctness and means yesterday's dashboard is stale. Continuing with a quality flag attached keeps things moving and requires every consumer to check the flag, which they will not. Choosing per dataset, before the first failure, is what separates a quality framework from a set of alerts nobody acts on.