Response structure

The preferred verification structure is claim_groups. Each group contains its claim, citations, domain groups, and optional verdict. The top-level citations, domain_groups, and verdict fields are legacy fields in the schema. New integrations should use the grouped structure.

totalResults counts citations across claims. thread_id can be used for conversation continuity. Source search shares this response schema but does not request verification analysis.

An omitted field, null value, false value, and zero have different meanings. Do not replace an absent verdict with supported, absent confidence with zero, or an unknown charge with a free operation.

The shared schema includes fields used by internal workflows, such as generated_prompts. Their presence in a response schema does not make internal request flags public parameters.

Every endpoint page shows its request fields, declared status codes, and response schema where one exists. Expand object fields for nested definitions. The OpenAPI file contains the complete published schemas.

Some operations do not yet publish a complete response schema. Those gaps are identified on their reference pages; illustrative examples are not substitutes for a schema.

Production rollout verified on 24 September 2026: the hosted API includes evidence policy 13 and the hosted MCP connector runs 1.9.0. This does not publish or upgrade the npm package installed by local clients. Inspect response metadata and the installed client version; saved responses retain their original policy and evidence time.

GET /api/v1/citations/{id} returns a saved record under data, including id, thread_id, created_at, and metadata when available. Preserve actual IDs; do not invent a thread ID, request ID or result URL when absent.

New stored results can include data.metadata.final_response. It holds the saved verification response and its evidence policy metadata. Reading it does not rerun verification. Preserve the stored verdict, source receipts and original evidence time. Historical citation storage may be a JSON-encoded bare source array or an object with citations or claim_groups; a missing stored verdict stays unavailable. Malformed storage is a data error, not an empty successful result.

The readback metadata is additive. Numeric fields that are unknown can be null. Update client types to preserve null separately from zero and false. Keep unknown additive receipt fields so a later backend does not lose provenance when your client stores the response.