Bounded candidate search behind the claim row's Link evidence picker. Returns up to 25 non-excluded evidence items from the intent's product, each with its trust tier and a short content excerpt, optionally excluding items already anchored to one claim. Never returns the product's full inventory.
Claim Trust Loop closeout (Sep 2026). Amendment 4 of the evidence trust contract forbids resolving claim backing from a browser-side copy of the product inventory, so the IntentSpec picker calls this route instead of listing all product evidence. Visibility is decided server-side: the intent must live in the caller's workspace and candidates are restricted to the intent's product; archived and dismissed rows never appear. Anchoring is not review: tier is reported so the picker can show what an item is, and turning an observed signal into backing remains the session-only atomic review action.
Path parameters
id path parameter.
Query parameters
Free-text search over content and source (case-insensitive, max 200 chars). PostgREST filter characters are stripped before matching.
Maximum candidates to return. Default 8; values above 25 are clamped to 25, never rejected.
A claim anchor key (objective, outcome:<id>, constraint:<n>, edgeCase:<id>). Items already anchored to that claim are excluded from the result.
Required for cookie-session auth; ignored for API keys, whose workspace is fixed.
Response
Candidates with trust tiers. tier is one of backing, unreviewed, assumption; excluded items are never returned.
Changes
Changed in 1 of the 14 revisions of this API.1
- ○
endpoint added
endpoint-added
- ○