nx_twinbench.nx

nx_twinbench.nx -- ★★★THE METRIC RULER, AND THE PRIMARY INSTRUMENT OF THE DIGITAL-TWIN PROGRAM. WHY THIS COMES FIRST. The program's governing law is capability = MIN(statistical, perceptual), and the perceptual term has always been an OPERATOR'S GLANCE -- no organ emits it, so every "we moved 175 -> X" has been unfalsifiable by construction. Aiming at a photoreal PLAUSIBLE human made that unavoidable: "does this look real" is a matter of judgement. Aiming at a 1:1 TWIN OF A SPECIFIC PERSON does not. "Is this surface within one millimetre of that person's face" is arithmetic. ★★THE TWIN TARGET MAKES MEASUREMENT EASIER, NOT HARDER -- it hands us a hard ruler the aesthetic target never could, and this organ is that ruler. WHAT IT MEASURES: the deviation of a candidate surface from a reference surface, as a distribution rather than one number -- mean, RMS, p95, and max (a one-sided Hausdorff). Reported in MILLIMETRES, because a surgeon does not plan in per-mille and a bench number that cannot be compared to a caliper is decoration. ★POINT-TO-TRIANGLE, NOT POINT-TO-VERTEX, AND THE DIFFERENCE IS NOT PEDANTRY. Nearest-VERTEX distance is the easy version and it is BIASED: it reports error proportional to the reference's tessellation, so a coarser reference scores worse for reasons that have nothing to do with the subject. A ruler whose reading depends on how finely you happened to triangulate the truth is not a ruler. So every query point is measured to the nearest point ON A TRIANGLE -- face interior, edge, or vertex, whichever is closest. ★A UNIFORM GRID, BECAUSE THE HONEST ALGORITHM MUST ALSO TERMINATE. Brute force is O(N*M): 35k vertices against 35k triangles is 1.2 billion point-triangle tests per comparison, and a ruler nobody runs is a ruler that does not exist. Reference triangles are bucketed into a uniform grid; a query expands its search radius ring by ring and STOPS ONLY when the next ring cannot contain anything closer than the best found so far. That last clause is what keeps it EXACT rather than approximate -- the grid changes the cost, never the answer, and T6 proves it against brute force on a small mesh. nx_twinbench cmp <candidate.nxmesh> <reference.nxmesh> [stature_mm] [layer] nx_twinbench selftest license_tier: ORIGINAL expect_exit: 0 No hw writes (Rule 26).