Recall-first MKU retrieval over the knowledge graph (MKF v0.1.2)
Retrieves adjudicated Medical Knowledge Units (MKU) for a query + patient EHR from the curated Neo4j graph: patient/query concept mapping, R1∪R2 retrieval, out-of-band safety sweep, LLM sufficiency judge, raw-chunk fallback, deterministic parent-doc enrichment, CROSS_REF reconciliation, and agent adjudication. Provenance-first and trace-rich (the eval surface).
Authorizations
Locus API key — paste the raw lsk_… value (no 'Bearer ' prefix).
Body
Free-text clinical question in French.
1"Quelle antibiothérapie pour cette infection urinaire ?"
Coarse patient EHR. Safety-signal facts (treatments, allergies, terrain) are mapped to concepts eagerly; used only in intersection with the query — never alone, except the safety sweep.
Response
Successful Response
Retained MKUs, ranked, with deterministic provenance.
Per-request eval data (logging is mandatory — spec §Logging).
Agent adjudication (optional). Absent when the agent is disabled — the structured retrieval below is the primary payload.
Ids of safety-relevant MKUs — addressed explicitly, never dropped.