A county page proves that a document exists. A working viewer proves that a viewer was observed. Neither establishes accepted parcel coverage.
Contract truth defines what must happen. Evidence truth records what a source says or a check establishes. Operational truth reports what the system actually did at a stated time.
A refresh policy belongs to a contract. A successful execution belongs to an observation. Whether a release is fresh is a derived evaluation of its vintage against the policy. Editing a status paragraph cannot make a pipeline healthy.
The evidence chain connects a user capability to a jurisdiction and layer, then a source contract, native schema, mapping, acquisition, validation, release, and product surface. Trace it in both directions: from a displayed attribute to its source, or from a schema change to every affected field and release.
The Census-based registry contains 115 county equivalents: 114 counties and independent St. Louis city. St. Louis County and St. Louis city are separate entries. The 2020 identity vintage does not describe parcel-data vintage or current source availability.
For statewide accepted parcel coverage, count distinct jurisdictions satisfying the declared parcel-release contract and divide by 115. Keep unresearched and blocked jurisdictions in that denominator. A pilot may report its own selected cohort denominator, clearly labeled; it cannot substitute for statewide coverage.
No statewide production numerator was observed for this corpus. Report unknown, rather than zero or a percentage derived from discovery records.
| Measure | Numerator and denominator | Required qualification |
|---|---|---|
| Research progress | Assessed jurisdictions / declared research scope | A research outcome, including a blocker, is not data delivery. |
| Accepted parcel coverage | Distinct accepted jurisdictions / 115 statewide | Declare layer, release scope, acceptance contract, and source vintages. |
| Extraction completeness | Validated in-scope records / expected in-scope records | Same source snapshot, filters, and documented count basis. |
| Attribute completeness | Valid required values / applicable records | Publish applicability rules and exclusions; non-null is not necessarily valid. |
| Layer completeness | Accepted required layers / declared required layers | Parcels cannot certify addresses, zoning, transfers, or permits. |
| Freshness | Data vintage compared with source-specific policy | Endpoint availability and successful fetches are separate observations. |
Unknown expected counts produce unknown extraction completeness. One response page, a sample, or a successful request cannot establish full extraction. An empty snapshot needs an explicit acceptance rule rather than division by zero. Competing sources cannot count a jurisdiction twice.
An accepted parcel release needs source identity, approval for intended use, geographic scope, schema and mapping revisions, acquisition evidence, validation results, lineage, freshness policy, and publication evidence. Numeric tolerances remain unresolved until reviewed; this documentation does not invent statewide geometry, count, or null-rate thresholds.
An exception needs scope, rationale, evidence, approver, review date, and user-facing meaning. A partial release must say which claims it cannot support. A derived association, such as an address-to-parcel match, must retain its method and uncertainty rather than appearing as a direct county attribute.
An eventual product can distinguish not collected, not provided, restricted, stale, failed validation, and no matching record when evidence supports that state. Its API and downloads need the same meanings. These are product requirements, not implemented interfaces.
The health reference defines how observations support operational status. The source-review guide defines the next action when evidence is incomplete.