Verdicts

Every /v1/evaluate response carries two top-level enums on purpose: You read both. status: ok + outcome: non_compliant is a clean evaluation that found a policy violation — block. status: degraded means at least one referenced source has no data in our store — treat with caution and consider retrying. status is driven exclusively by whether each referenced source has ever been ingested (we check dataset_references[].collected_at_latest internally). A transient upstream failure during a single request does not flip status to degraded as long as we have prior data for the source — you’ll see dataset_references[].collected_at_latest carrying the older timestamp, and it’s your call whether that data is fresh enough for your use case.

“No intersections in the queried area” is NOT degraded. A source with rows in our store but zero overlapping your input geometry stays status: ok and the relevant check returns zero hits. Degraded only signals “we have nothing at all for this source”.

​

Outcome semantics

​

Severity

Each check declares one of five severities — critical, high, medium, low, info. The default policy maps them to outcomes roughly like this:

• critical / high → triggering pushes the outcome to non_compliant.
• medium / low → triggering pushes the outcome to warning (if not already non_compliant).
• info → never affects outcome; surfaces as a triggered check for reporting.

Rulesets can also declare an explicit verdict_policy to override the default mapping. See the authoring guide.

​

Evidence

Each triggered check carries up to 20 matching features in evidence[] — the actual items that caused the check to fire. Use them to render explanations to your users (“polygon overlaps the Yanomami Indigenous Territory by 0.42 km²”) instead of opaque pass/fail flags. Each check also reports total_matched — when this exceeds evidence.length, the UI should indicate truncation.