Data quality score
How much of your data passes the rules you have agreed it must meet.
01 What it is
Why anyone looks at this number
How much of your data passes the rules you have agreed it must meet.
Every metric on this site inherits the quality of the data underneath it. A governed number built on unvalidated data is a confident wrong answer.
02 The formula
How it is worked out
Data quality score = weighted pass rate across completeness, validity, uniqueness, timeliness and consistency rules grain : daily × dataset × domain unit : percentage source : The data platform's quality engine, run against the source systems
Weight the rules by business impact. An unweighted score treats a null in a comment field the same as a duplicate customer, and then nobody believes the score.
03 Worked example
The same number, with real inputs
| Rules passing | 1,412 |
| Rules run | 1,530 |
| Critical rules failing | 9 |
| Calculation | 1,412 ÷ 1,530 |
| Result | 92.3%, with 9 critical failures |
The headline looks healthy; the nine critical failures are the story. Score by severity, or the important failures hide inside a comfortable average.
04 What moves it
Four things that actually change this number
Rule coverage
You only score what you test.
Source system change
A new field or a changed validation upstream.
Manual entry points
Free-text and spreadsheet stages, which is where most failures start.
Ownership
Whether a named steward actually fixes the failures.
05 Where the number lives
The system, the record and the fields
| System of record | Key record | Fields you need |
|---|---|---|
| The data platform's quality engine, run against the source systems | Dataset, Column and Rule | rule_id, dimension, rows_tested, rows_failed, severity, owner |
Weight the rules by business impact. An unweighted score treats a null in a comment field the same as a duplicate customer, and then nobody believes the score.
06 How it goes wrong
Three ways this metric misleads people
Unweighted scoring
Trivial rules outnumber critical ones, so the score barely moves when something serious breaks.
Fix: Weight by severity and report critical failures separately.Testing the warehouse only
Failures found downstream have already reached reports.
Fix: Test at ingestion, where the fix is cheapest.No owner per rule
Failures are reported forever and never resolved.
Fix: Every rule carries a named steward and an expected resolution time.08 Questions
Frequently asked
What are the standard data quality dimensions?
Completeness, validity, uniqueness, timeliness, consistency and accuracy. Accuracy is the hardest because it needs a real-world reference, not just a rule.
Where should quality be tested?
As close to the source as possible. A rule that fires at ingestion prevents a wrong number; the same rule in a dashboard only explains one.
One definition, everywhere it is used
SCIKIQ stores this metric once and serves it to every dashboard, board pack and agent that asks.