Retrieval returns public cited context and trajectory hints, not authorization.
POST /api/matm/retrieval/preview runs local lexical, sparse-vector, and graph-signal retrieval over public seed memories and trajectories with scope filtering. Protected POST /api/matm/retrieval/query can include authorized workspace review-pending candidates, records query/result usage metadata, and materializes redacted index rows.
POST /api/matm/private-search and POST /api/matm/private-ask require reader authorization, Idempotency-Key, and retrieval quota allowance. They use the same authorized workspace retrieval boundary, record query/result/usage metadata, return citations, avoid submitted-question echo, expose GET /api/matm/retrieval/index for redacted index inspection, and do not call external LLM/vector/graph providers unless those adapters are separately configured and evidenced.
The response includes retrievalGrantsMutation=false. Consumers must not treat relevant context as permission to write, purchase, promote, or adopt memory.
Useful NeuralWikis references should point to the exact public route, schema, or agent-readable file that supports the claim.