nx_nxa_play_gate.nx

nx_nxa_play_gate.nx -- THE POSE PLAYER GATE: skinning is JOINT-DRIVEN, deterministic, bounded at identity, and refuses an unrigged asset by name -- each proven on REAL corpus artifacts. Subject: ./nx_nxa_play.elf (the serving-root binary; e2e fork via nx_gatekit_lib's proven capture, never a hand-rolled pipe -- a v1 of a sibling gate deadlocked in pipe_wait for 50+ minutes on exactly that mistake, and the banked law is that fixtures and subprocess come from nx_gatekit_lib). THE ANTI-VACUITY TOOTH IS SPREAD, NOT MOTION. "The mesh moved" is passed by a global translate and by uniform jitter -- neither is skinning. A joint-driven deformation moves different vertices by DIFFERENT amounts (a shoulder swings the hand far, the hip barely at all), so max-min displacement is strictly positive; a whole-mesh shove makes every vertex travel the same distance and collapses that spread to ZERO. This gate therefore asserts SPREAD, which the trivial wrong implementation cannot produce. IDENTITY IS BOUNDED, NOT EXACT, AND THE BOUND IS DERIVED. Applying the identity pose does NOT reproduce bind pose bit-for-bit: nx_skeleton blends in fx256 and truncates once per influence in `w * (...) / 256`, plus once in the matvec, so the worst-case residual is INFLUENCES + 1 = 5 model units. Measured across all 16 corpus rigs (14,164 to 423,919 verts) the actual floor is a stable 3. The tooth asserts the DERIVED bound of 5 and PRINTS the observed value, so a regression that pushes residue past truncation noise fails while honest rounding passes. This floor is also the CONTROL for any future twist/volume claim: a real LBS artifact must exceed it to be real at all.