What quality actually means.
WHAT THE DIMENSIONS ARE
- Accuracy: values reflect reality
- Completeness: nothing missing
- Consistency: agreement across systems
- Timeliness: available when needed
- Validity: conforming to defined rules
- Uniqueness: no unintended duplicates
WHAT ACCURACY IS HARDEST TO TEST
Because it requires comparison with reality, not with rules.
WHAT PROXIES EXIST
Reconciliation against a system of record Comparison with independently known figures Review by people who know the domain
WHAT COMPLETENESS PROBLEMS LOOK LIKE
Missing rows, which are invisible unless expected counts are known.
WHAT TO ESTABLISH
Expected volumes, so absence is detectable.
WHAT CONSISTENCY PROBLEMS LOOK LIKE
Two reports disagreeing, usually traced to different definitions or filters.
WHAT VALIDITY TESTS CATCH
Values outside permitted sets or ranges, and malformed formats.
WHAT UNIQUENESS FAILURES CAUSE
Inflated aggregates, which look plausible.
WHAT TO MEASURE AND PUBLISH
Quality metrics per dataset.
WHY PUBLISH
So consumers know what they are relying on.
WHAT TO AGREE
Thresholds at which data should not be published at all.
