nx_finger_count.nx

nx_finger_count.nx -- Tier 6 anatomical-coherence finger-counter. THE canonical generative-AI failure mode: six fingers per hand, or three, or seven, or two fingers fused into a paddle. Stable Diffusion / Flux / Midjourney / DALL-E all produce this with alarming frequency because they have no anatomical prior. Substrate's bits-up answer: count fingers DETERMINISTICALLY via horizontal scan-line crossings on a binary hand mask. Method: Input: binary hand mask (0 = bg, 1 = hand), w x h, plus a scan_y row index that cuts horizontally THROUGH the fingers (above the palm, below the fingertips). Caller determines scan_y from hand-orientation upstream. For x = 0..w-1: track run-length of consecutive 1 pixels along scan_y. Each contiguous run >= min_run_width is one finger. verdict by run count vs anatomical norm (5 per hand): 5 -> OK 6+ -> EXTRA (the famous AI failure) 4 -> MISSING (or fused) 3 or less -> FUSED / OCCLUSION NOISE detected (many tiny runs) -> NOISY Caller responsibility: provide a HAND-segmented binary mask + a VALID scan_y. Substrate provides the deterministic counter. Reference: Suk/Park 2010 "Hand gesture recognition based on digital image processing using MATLAB" -- scan-line approach for finger counting in segmented hand regions. Convex-hull-defect approach (Hawley/Eichmann 2014) is more robust to fingertip occlusion; queued for nx_finger_count_v2. genealogy_id: suk_park_2010_scanline lineage_id: substrate_anatomy_coherence_v1