Name the four dimensions of data quality

At a glance

Learning objectives

• Name the four dimensions of data quality — freshness, completeness, validity, consistency
• Map each dimension to a concrete signal — SLA, null rate, range, uniqueness
• Rank failure modes by blast radius before writing any test

What you need

• No installation — this lesson reads registry artifacts, it does not run them
• The cited artifacts — the data-contract skill, the data-quality reviewer agent, the quarantine starter

Steps

1. The SLA part of a contract states freshness and completeness — the first two dimensions, with numbers. (from skill-data-contract-author )
2. Semantics is next — what each field means, units, nullability — the validity a consumer was promised. (from skill-data-contract-author )
3. Profile key columns (null rate, uniqueness, ranges) — the measurements that turn each dimension into a check. (from agent-data-quality-reviewer )
4. Rank failure modes by blast radius before writing a single test. (from agent-data-quality-reviewer )
5. The starter's rule vocabulary covers the dimensions — not_null, unique, range, enum. (from project-pipeline-dq-quarantine )
6. Monitor the three signals that watch the dimensions in production — expectation pass rate, quarantine row counts, freshness. (from runbook-data-quality-failure )

Check yourself

• A table loads six hours late — freshness; a promised column is 30% null — completeness; profiling key columns (null rate, uniqueness, ranges) surfaces both.
• The SLA part of the contract names freshness and completeness, and the semantics part names units and nullability.
• Failure modes are ranked by blast radius, and the minimum test set follows from that ranking.

Worked example

1. Owner (a team, with an on-call).
2. Schema (typed, versioned).
3. Semantics (what each field *means*, units, nullability).
4. SLA (freshness, completeness).
5. Breaking-change policy (notice period, deprecation path).
6. Enforcement (which CI test fails on violation).

Profile key columns (null rate, uniqueness, ranges).
Rank failure modes by blast radius.
Propose the minimum test set.
Emit a quarantine path for rows that fail.

contract/orders.yml  # rules: not_null, unique, range, enum

Sourced from

• Data Contract Author
• Data Quality Reviewer
• Pipeline → DQ → quarantine → publish
• Data quality failure
• Contracts as a Quality Gate

Security scan

Last sweep 2026-10-24: 0 finding(s), 0 marked as examples.

Common gotchas

• A contract with no enforcement is documentation. Documentation is not a contract.
• Producers change schemas and semantics with no notice; consumers break in production.
• Profiling runs against a provided sample or a read-only replica only; never writes to the source.

Production considerations

• Monitor expectation pass rate, quarantine row counts, freshness.
• Propose the minimum test set — the dimensions ranked by blast radius, not every column.

Diagram

the four dimensions
   |
   |- freshness    --> how recently the data landed (SLA)
   |
   |- completeness --> null rate on promised columns
   |
   |- validity     --> ranges, enums, types, units
   |
   |- consistency  --> uniqueness on the grain key, stable semantics
   |
   |- profile key columns (null rate, uniqueness, ranges)
   |
   |- rank failure modes by blast radius --> minimum test set

Registry stats