Skip to content
Data Quality Observatory

The measurement system must be observable too

Bay Evidence does not treat datasets as context-free facts. Every important metric should expose its provenance, coverage, revision history, and comparability constraints.

Minimum public metadata

Source date
Publication date
Retrieval date
Geography
Population universe
Methodology version
Revision history
Known exclusions
Suppression rules
Missingness
Comparability warning
Review status
Verification status

Quality and comparability states

CURRENT
Verified and methodologically comparable for the intended analysis.
CURRENT_PROVISIONAL
Recent data available, but revision or methodological clarification remains possible.
STALE
Historically useful, but not suitable for claims about current conditions.
METHOD_BREAK
A change in method, timing, definitions, or coverage compromises direct comparison.
PARTIAL_COVERAGE
The data universe excludes meaningful parts of the target population or geography.
CONFLICTED
Credible sources disagree materially and the discrepancy is not yet reconciled.
SOURCE_NEEDED
A claim or figure is circulating without an adequate authoritative source.
UNDER_REVIEW
Discovered or ingested but not cleared for headline or analytical use.

Comparability is not automatic

Two jurisdictions can publish measures with the same label while using different universes, geographies, collection windows, deduplication rules, or operational definitions. A trend can also break when one jurisdiction changes enumeration timing, survey approach, administrative coverage, or eligibility rules. Bay Evidence should make those method breaks visible at the point of use.

Harmonization is documented transformation

Harmonizing data does not mean forcing every source into one false equivalence. It means preserving original values, documenting transformations, maintaining versioned definitions, and exposing what remains incomparable. The original source should always remain recoverable.