Four indicators that matter
Teams often try to measure data quality with a composite score that nobody trusts. The usable version is smaller and simpler.
1. Completeness
What share of expected observations actually arrived? If a daily series has 252 expected observations in a year, how many landed? Gaps should be visible, not hidden.
2. Range validity
Are values within plausible bounds? An equity price of negative ten dollars, a yield of 500%, or a volume of zero on a trading day should all be flagged automatically.
3. Cross-source reconciliation
Where a second source exists, how often do they agree within a threshold? A small, stable disagreement is fine; a sudden widening disagreement is a signal.
4. Duplication and freshness
Are there duplicate rows, and is the data as fresh as expected? A daily job that silently stopped running is the most common failure we see in internal pipelines.
These four checks, run on every refresh and surfaced as a short status page, tell you whether a dataset is usable. Anything more elaborate is usually theatre.