SCRIMED CODE pt. 4
Hybrid Retrieval Engine
SCRIMED Hybrid Retrieval is a synthetic/no-PHI retrieval architecture layer. Clinical settings require citations and evidence; the system must not answer clinical questions without evidence and does not authorize diagnosis, treatment, prescribing, EHR writeback, or payer submission.
Unified Ranking
0.32*bm25 + 0.38*vector + ontology + graph + trustTier
rank 1
Reviewed clinical guideline summary placeholder
Do not answer clinical questions without evidence. Cite sources, expose uncertainty, and route high-risk content to human review.
reviewed
- Unified score: 0.844
- BM25: 0.78; vector: 0.8
- Ontology boost: 0.1; graph boost: 0.09
- Audit hash: scrimed-intel-1fcbd15b
rank 2
FHIR Observation semantic mapping fixture
Do not answer clinical questions without evidence. Cite sources, expose uncertainty, and route high-risk content to human review.
synthetic_fixture
- Unified score: 0.803
- BM25: 0.7; vector: 0.84
- Ontology boost: 0.12; graph boost: 0.1
- Audit hash: scrimed-intel-aafcec45
rank 3
Synthetic prior authorization policy criteria
Do not answer clinical questions without evidence. Cite sources, expose uncertainty, and route high-risk content to human review.
internal_policy
- Unified score: 0.771
- BM25: 0.82; vector: 0.76
- Ontology boost: 0.08; graph boost: 0.06
- Audit hash: scrimed-intel-956ab0f2