A quality score needs a denominator

DataGovernance.ca editorial · 13 October 2026 · Implementation guidance

Name the decision and the data element

Choose a field or metric because it affects something important, not because it is easy to count. An email-format check may be straightforward, but it does not prove the record belongs to the right person. A report can be internally consistent and still use the wrong definition. Write a plain-language statement of what the test can and cannot establish.

Specify the rule completely

For each check, record the population, reporting period, accepted values, exclusions, pass condition and threshold. Define the numerator and denominator. For example, a completeness measure could be the number of scoped records with a required owner identifier divided by the number of scoped records that require one. Keep the actual threshold a business decision; do not copy an arbitrary percentage into every field.

Make exceptions visible

Some exceptions are legitimate; others reveal a broken process. Identify who can approve one, how long it lasts and when it is reviewed. Keep excluded records visible in the results so the rate cannot improve merely because difficult records were removed. Record changes in the test definition before comparing periods. A better percentage with a different population may not represent an improvement.

Connect the measure to a response

The useful part is what happens when the check fails. Assign an owner, a triage path and a method for distinguishing data repair from root-cause correction. Agree how the business will operate while an issue remains open. Begin with a small number of well-defined checks and a working response routine before expanding the dashboard.