Record feedback
Persist an agent or evaluation feedback event.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Headers
Optional client session identifier.
Body
What happened: result_helpful / result_not_helpful (the result did or did not advance the task), passage_used (a passage was used or cited), read_abandoned (opened but abandoned), duplicate_result, stale_result (outdated content), spam_or_low_quality, missing_expected_source (a source you expected did not appear), unsafe_or_policy_issue.
result_helpful, result_not_helpful, passage_used, read_abandoned, duplicate_result, stale_result, spam_or_low_quality, missing_expected_source, unsafe_or_policy_issue Optional calling-agent context for slicing feedback in evaluation.
Document the event concerns, from the result's doc_id.
Free-text context, recorded for human review and offline evaluation.
Specific passage the event concerns, from passages[].passage_id; most useful with passage_used.
Query text the event relates to, useful when no search_id is available.
One-based position the result had in the ranked list.
x >= 1The search_id returned by /v1/search. Ties feedback to the exact ranked list that produced the result, so ranking learns from the right context.
Session the event belongs to (UUID), matching session_id from earlier responses.
Response
Feedback accepted.