What are Data Quality Dimensions?

At scale, bad data drives bad decisions at scale. A pricing model trained on corrupted data can lose millions per day. A recommendation engine fed incomplete events shows users irrelevant content. Data quality dimensions transform vague complaints like "data is bad" into explicit, measurable properties you can monitor and enforce. Accuracy: Is it Correct? Accuracy measures correctness relative to the real world. If an orders table shows a user paid $9.99 but the payment processor charged $19.99, that data is inaccurate. This goes beyond format validation. A value can be syntactically valid but semantically wrong. Typical checks include comparing derived data against trusted sources, or validating realistic ranges: latitude between negative 90 and 90, user age between 13 and 120. Completeness: Is Everything Present? Completeness measures whether all expected data arrived. A dataset can be accurate but incomplete. You might have all orders that did arrive perfectly recorded, but be missing 1 percent due to Kafka lag or a broken ingestion job. Completeness is quantified as a percentage: for example, 99.9 percent of expected events for a given hour actually arrived in the warehouse. Consistency: Does it Agree With Itself? Consistency measures whether data aligns across different views or systems. Within a table, this covers constraints like unique identifiers and referential integrity. Across systems, it means user state in the data warehouse matches the source systems within expected lag. You can tolerate some delay, but not permanent contradictions like an order marked "refunded" in one table and "completed" in another forever.