nx_meshgap_gate.nx

nx_meshgap_gate.nx -- THE COHERENCE TOOTH: does a shipped character hold together in space? WHY (measured 2026-08-23). The models page card FEMALE DARK KNIGHT renders as FRAGMENTED GEOMETRY -- hair floating far above the body, blobs below -- while EVERY NUMBER ON ITS CARD IS GREEN (33 geometry nodes merged, rig 113 joints, 3729 clusters = 33 x 113, albedo carried). The arithmetic closed and the picture was broken. Third time in one day the eye beat the teeth, and always for the same reason: every tooth measured a COUNT or a SUM, and none measured whether the thing HOLDS TOGETHER IN SPACE. THE BOUND IS DERIVED FROM A MEASURED POPULATION, NOT PICKED. All four ingested FBX donors were measured with nx_meshgap and the ranking matched the eye exactly, in order: toon3d8 max_gap_permil = 8 renders coherent dark_witch max_gap_permil = 20 renders coherent paladin max_gap_permil = 101 one boot detached (mild) dark_knight max_gap_permil = 298 severely fragmented The coherent population tops out at 20; the fragmented population starts at 101. Any bound in between separates them, so rather than pick a point in that range the bound is the GEOMETRIC MEAN of the two boundary values -- sqrt(20 * 101) = 44.9 -> 45 -- which sits at an equal RATIO margin (2.25x) above the worst coherent donor and below the best fragmented one. Ratio-symmetric is the right symmetry here because the measure is itself a ratio (permil of the axis's own span). AND THE BOUND CANNOT SILENTLY GO STALE: T5 re-derives the separation from the live corpus on every run and FAILS if the two populations ever stop being separated by this bound. A frozen number that nobody re-checks is how a threshold becomes folklore. THE NEG-CONTROLS ARE REAL RECORDED DEFECTS, NOT SYNTHETIC MUTANTS: dark_knight and paladin are genuinely fragmented assets sitting in the corpus right now. Biting against the estate's own history is stronger than any mutant, because it proves the tooth catches what actually happened.