Data Quality Dimensions
The standard axes along which data quality is measured, which turn a vague complaint into a testable assertion.
"The data is bad" is not actionable. The dimensions exist to force specificity, and each maps to a different kind of check and a different remedy.
Completeness — are values present where required, and did all expected records arrive. Accuracy — does the value match reality, which is the only dimension that cannot be checked without an external reference and is therefore the one most often assumed rather than measured. Consistency — do related values agree within and across systems. Timeliness — is it available when needed and does it reflect a recent enough state. Validity — does it conform to its defined format, type and range. Uniqueness — does each real entity appear once.
The architectural point is that these are checked at different places. Validity and uniqueness are cheap and belong at ingestion, where the failure can be rejected. Completeness needs an expectation of what should have arrived, which means the pipeline must know its expected volume rather than processing whatever appears. Consistency requires cross-system reconciliation, usually a scheduled control rather than an inline check.
The practice that changes outcomes is treating quality checks as assertions in the pipeline that can fail the run, rather than as a monitoring dashboard nobody reads. A dashboard reports the corruption; an assertion prevents it propagating.