nx_doc_classify.nx

nx_doc_classify.nx -- R3 of THE NISHI DOCUMENT-INTELLIGENCE arc: CLASSIFY a document by TYPE (medical bill / bank statement / EOB / letter / ...) so the pipeline ROUTES it to the right extractor before R1. The "organizing" leg. DATA-DRIVEN BY CONSTRUCTION (#6/#11/#25): each type is a PACK of (keyword, weight) signals passed as DATA (grounded on the doc_cls_* research corpus + a sovereign rule-pack store in R6 style). Classification = score each type by summing the weights of its signals present in the text; the highest-scoring type above a threshold wins, else "unknown" (-1). Deterministic, exact integer scores, no ML dependency, no hardware writes. Adding a type or tuning a weight is DATA, never a code change. (v1 substring match is case-sensitive; case-folding is the noted refinement.) license_tier: ORIGINAL