nx_doc_scan.nx

nx_doc_scan.nx -- R4 of THE NISHI DOCUMENT-INTELLIGENCE arc: the SCAN / CAPTURE leg (paper/image -> characters). The honest first rung of OCR, built sovereign rung-up (NOT a waved "general OCR"): (1) DECODE a standard scanned-image format -- NetPBM P1 (ASCII bitmap, "P1\n<w> <h>\n<0/1 grid>") -> a pixel grid. A real image-file decoder. (2) RECOGNIZE digits by TEMPLATE MATCHING (the classic fixed-font OCR technique): each glyph cell is compared against a sovereign 3x5 reference font (DATA), nearest by Hamming distance wins; the distance IS the confidence (a blank/garbage cell matches at a large distance = low confidence). This turns a scanned bitmap (e.g. an account number / amount) into a structured digit string. Honest scope: fixed 3x5 digit font + fixed-pitch cells; variable fonts / proportional layout / full alphabets are later rungs grounded on the doc_scan_* / doc_read_layout corpus. No hardware writes, no floats. license_tier: ORIGINAL