// MODULE
axre-graph is the BFO-2020-typed clinical knowledge hypergraph — anatomical sites, findings, disorders, mechanisms, procedures, and substances, with the typed relations between them, grounded to SNOMED CT / UMLS / MONDO / HPO and curated for proof use. It holds the type-level background knowledge that axre-reason draws on; it carries no case/instance data.
A BFO-2020-typed, fully provenance-scoped hypergraph in which clinical knowledge exists not as a collection of facts but as a proof-capable mechanism net. Every node is a typed entity anchor (structure, function, disposition, process, disorder, substance, finding), grounded against SNOMED CT / UMLS / MONDO / HPO and consistently linked through the loaded ontologies (BFO, RO, MONDO, HPO, UBERON, CL, ChEBI, GO, NBO, MFOEM …) — no loose strings; IDs drawn end-to-end from the live graph.
The backbone is the etiology-agnostic localization spine: deficit pattern → structure → function → deficit, with function as the occurrent pivot carrying both loss tails (lesion → deficit) and gain tails (stimulation → ictal symptom), discriminated via perturbation_type. Etiology attaches through shared structures, never as a localization anchor.
On top sit the fully integrated domain arcs — neurology (brainstem, peripheral-somatic, cortical/subcortical, visual pathways) and psychiatry (ICD-11 Ch. 06 as a curated gate/bridge structure) — plus the PNEI layer coupling psycho-neuro-endocrine-immune mechanisms across systems. The drug layer (RxNorm, DrugCentral, receptor edges) attaches etiologically, never on the proof path.
Growth only through the fixed 32 predicates; every new edge provenance-partitioned, every reasoner-relevant attribute living in curated YAML. The result: a graph over which axre-reason can run strict-domination proofs yielding the four verdicts (proved / refuted / insufficient_input / abstained) — fully auditable, non-probabilistic, every step traceable to source and rule.
Optional, once Rail-A closes: a literature-fed expansion layer that admits content only past a ≥90% precision gate — an enriched but cleanly separated provenance layer, never on the core proof path.