Data Quality Rule

How it works in practice

An analytics team runs a model that computes daily active users. They attach four rules to it: user_id is unique, event_date is not null, the count stays within 50 percent of the trailing seven-day average, and the source table was loaded in the last two hours. One morning the freshness rule fails, the count rule passes, and the model is held back before it can publish a stale number to leadership. Tools like dbt tests and Great Expectations make rules like these declarative and run them on every pipeline execution.

Data quality rule vs. its neighbours

• Data contract🔣Data ContractComplianceA data-sharing contractView reference → is the broad agreement between producer and consumer. A quality rule is one enforceable clause that may sit inside a contract, but rules also live independently as pipeline tests.
• Data lineage🌳Data LineageData & AnalyticsA record of data origin and transformationsView reference → finds where a problem came from. A quality rule detects that a problem exists in the first place. A failing rule plus lineage is the standard root-cause pairing.
• Metric🔢MetricStrategyA unified metric that measures progress, health, or behaviour across the productView reference → is the business number being computed. A quality rule is the guard on the inputs and the output of that number, which is why the canonical edge runs from metric to rule.

The six widely cited dimensions, formalised by Wang and Strong in 1996, give a vocabulary for what rules check: accuracy, completeness, consistency, validity, timeliness, and uniqueness.

In the graph

In the Unified Product Graph, a data quality rule sits in the data and analytics region and connects most directly to the things it protects. A metric is validated by a rule (metric_validated_by_data_quality_ruleMetricvalidated byData Quality Rulehierarchy), and rules associate with the data_source🛢️Data SourceData & AnalyticsA data source or integrationView reference → they test and the data_lineage🌳Data LineageData & AnalyticsA record of data origin and transformationsView reference → that traces failures. That metric edge is the meaningful one: it turns an abstract assertion into a guarantee attached to a specific business number, so you can ask which metrics are guarded and which are flying blind.