Six Dimensions of Data Quality

The six axes you actually score

1. Accuracy — does the value reflect reality?
2. Completeness — are required fields present?
3. Consistency — does the same fact agree across systems?
4. Timeliness — is the value fresh enough for its use?
5. Uniqueness — exactly one row per real-world entity?
6. Validity — does the value satisfy its declared format / range?

A practical data-quality programme scores datasets on each axis, sets a target per axis per dataset (perfection is uneconomic), and instruments alerts on drift.

Quality is a control loop, not a score

A dimension only becomes operational when it has five parts: a measurement (SQL or expectation), a threshold tied to a consumer use case, an owner, an incident route, and a remediation decision. For example, customer_id uniqueness may require 100% for billing but only 99.5% for exploratory analysis. The failed check must either block publication, quarantine rows, degrade a trust badge, or open an issue. A dashboard with no consequence is observability, not quality management. Measure at the producer boundary where defects are cheapest to stop, and again at critical consumer boundaries where meaning can drift.

Think of a hospital chart:

• Accuracy = the right patient's vitals were entered.
• Completeness = no blank allergy field.
• Consistency = the same blood-pressure reading on the chart and the monitor.
• Timeliness = today's vitals, not last week's.
• Uniqueness = one chart per patient, not three with name variants.
• Validity = the temperature is plausible (35–42 °C), not 410.

Bad scores on any one axis can kill a patient — same idea for a business decision.