DATA QUALITY METRICS SCORECARD
Reader package prepared October 13, 2026.

Files
data-quality-scorecard-template.csv: blank fields for your own workflow.
data-quality-scorecard-worked-example.csv: static fictional billing example.

Open the CSVs in your spreadsheet application. They contain no macros,
external connections or automatic formulas. Calculate your measures and
record the results yourself. Do not overwrite the worked example as evidence
of a real run. The file contains assumed counts, not a raw event dataset.

WORKED EXAMPLE
Expected population: 1,000 distinct billable events in one closed window.
Delivered population: 990 rows, 985 distinct expected IDs, 5 excess copies.
All delivered event IDs are nonblank. There are no unexpected IDs.
Missing expected events: 1,000 - 985 = 15.
All 990 delivered rows meet the three stated field/reference/quantity rules.
Accuracy, monetary exposure and freshness were not measured.
The expected ledger and reference are fictional teaching assumptions.

CHECK THE CALCULATIONS
Required-field completeness: 990 / 990 * 100 = 100.00%.
Expected-event coverage: 985 / 1000 * 100 = 98.50%.
Excess duplicate rate: 5 / 990 * 100 = 0.50505...%, rounded to 0.51%.
985 + 15 = 1,000 expected distinct events.
985 + 5 = 990 delivered occurrences.
Do not average these rates into an overall release score.

USING THE TEMPLATE
1. Name the decision and dataset grain.
2. Agree the eligible population, window, timezone and exclusions.
3. Document the reference evidence and its limitations.
4. Name one metric per row and define numerator and denominator.
5. Record counts and rate. A zero denominator means not evaluated.
6. Agree thresholds from business consequences, not a universal percentage.
7. Name investigation and release approval roles.
8. Rehearse a consequential failure and document the recovery decision.

Late events, replay across prior windows and corrections require explicit
policies. This example only measures duplicates inside one window.
An unknown result must not be recorded as a passing result.
Do not upload private customer records into an AI tool to fill the template.
The template only needs metric definitions, counts and decision notes.

Read the article: https://shyamuthaman.com/data-quality-metrics/
Planned article URL, confirm after publication.
Related experiment: https://shyamuthaman.com/data-contract-validation-experiment/
Book: https://mybook.to/theDataDrivenExec_SU
