nx_meshvalid_lib.nx

nx_meshvalid_lib.nx -- THE MESH VALIDITY REFEREE: it REFUSES a broken asset instead of shipping it. /compare/dcc DC3, and it is the tooth the retopology and sculpt rungs both bind their claims to. WHY THIS AND NOT AN OVERLAY. Blender, ZBrush and Maya all SHOW mesh problems to an artist in a viewport overlay; none of them REFUSES on one, because a person is always there to look. An estate that emits assets from rules with no person in the loop needs the opposite: a named count, the offending element index, and a non-zero verdict. nx_spendgate found no incumbent -- nx_meshgen BUILDS watertight meshes by construction (surface nets) and nx_step_tess2 tessellates with hole bridging, but nothing in the estate VALIDATES a mesh it did not itself produce, which is exactly the case an importer creates. EVERY DEFECT IS ITS OWN NAMED CLASS, never a single boolean. A compound assertion that will not name its failing conjunct is a false-alarm generator: the reader always guesses the alarming one. Six classes, each with a count AND a first-offender index, and a manifold verdict composed from them. COMPLEXITY IS DECLARED, NOT DISCOVERED. Edge and vertex identity both go through an open-addressing hash table whose capacity is DERIVED from the input size (next power of two at or above four times the element count, a load factor at or below one quarter), so the check is linear in triangles rather than the quadratic pairwise scan the naive version would be. There is no fixed table size to outgrow and no cap to silently hit. All integer: the zero-area test is an exact cross product, never a float epsilon, so DEGENERATE here means exactly degenerate rather than nearly so. 100% sovereign. No hardware writes (Rule 26). license_tier: ORIGINAL