// MODULE
Clinical NLP pipeline for axiomatic extraction. The first stage of the AXRE environment. axre-extract reads medical exam–style case vignettes and turns them into a typed, SNOMED CT / UMLS–grounded clinical hypergraph — every finding, anatomical site, procedure, medication, lab result, and candidate diagnosis, with negation, assertion status, and the relations between them. Each case is emitted as a sealed, self-contained hypergraph artifact (reasoner_input.json); downstream AXRE modules consume that artifact. ⚙️ Deterministic, not LLM-dependent — same input → same graph; the LLM is optional and off by default, contributing zero graph structure (concepts, relations, question types). 🔗 Typed predicate-logic hypergraphs — clinical facts as a queryable typed hypergraph artifact, not opaque embeddings or flat triples. 🩺 Grounded to UMLS/SNOMED — every concept resolved to a standard clinical CUI (MedCAT · SapBERT · SciSpaCy), not a free-text label. 🛑 Hard-fail, no silent degradation — a missing or degraded component aborts loudly and quarantines the case; it never emits a quietly corrupted graph. 🧱 Extraction ≠ inference — surface clinical facts only; diagnosis is left to downstream reasoning, never baked into the graph. 🔒 Answer-blind by construction — for exam-derived inputs the gold answer never enters the graph (structural firewall).