nx_doc_extract.nx

nx_doc_extract.nx -- R1 of THE NISHI DOCUMENT-INTELLIGENCE arc: READ a document's text into STRUCTURED DATA. The "consuming documents -> structured data" leg (the other leg, making documents, is nx_fin_dispute / the emitters). Sovereign + deterministic: a tolerant line/field parser (no OCR/ML dependency) that turns an itemized medical-bill statement into structured line-items (date, CPT code, units, billed cents) that flow straight into nx_fin_audit -- so a scanned/typed bill becomes auditable structured data with ZERO manual entry. This is the format-specific deterministic rung; the research corpus (knowledge/fetched/ doc_*.raw: layout-analysis, IE, NER, classification) grounds the roadmap to arbitrary-layout + classified extraction. A line is a line-item iff it begins with a MM/DD/YYYY date AND carries a 5-digit code AND a $amount; headers/totals/blank lines are skipped by construction. No floats (exact cents), no hardware writes. license_tier: ORIGINAL