Context graphs¶
The optional context service stores evidence slices, assembles a bounded
retrieval context, and records provenance. It is available only when configured.
GET /context/health returns 200 when the service is available and 503 when
it is disabled. A healthy main server does not imply that context retrieval is
enabled.
| Method and path | Operation |
|---|---|
POST /context/ingest |
Write context slices and provenance through the shared RDF update path. |
POST /context/assemble |
Retrieve and rank permitted slices into a context envelope. |
POST /context/retract |
Remove selected slices or the slices of one tenant/domain source graph. |
GET /context/runs/{id} |
Read recorded context-run information. |
With authentication enabled, the verified actor supplies tenant and SID scope. Client-controlled SID headers cannot replace that identity. Access filtering applies before a result can enter the returned context.
Retraction¶
A retract request supplies domain, optional tenant, and exactly one target:
up to 500 slice_ids or one source_graph. A source graph must match the selected tenant and
domain, such as https://id.indentia.ai/sources/acme/research.
Retraction removes the selected slices from canonical facts and the hybrid search projection. A source-graph retraction removes slices, not unrelated triples that happen to share the graph. Provenance and access checks remain part of the operation. This is a write endpoint, not a client-side result filter.
Asset status and ranking¶
CONTEXT_STATUS_WEIGHTING opts in to status-aware ranking. Status is canonical
governance data, changed through governed commits. Ranking weights do not grant
access, replace source evidence, or make a draft fact an approved decision.
Context consumers must retain the returned evidence and scope. A retrieved slice supports a proposed answer; it does not authorize a downstream write or external action. Use Governed commits when a resulting mutation needs policy proof and a recoverable receipt.