nx_gltf_anim_gate.nx

nx_gltf_anim_gate.nx -- export the being as a SKINNED, WALKING .glb. Mesh the skin, bind each vertex to its nearest part's deepest FK bone, export with a walk-cycle animation, self-verify the container + names. Writes knowledge/nishi_being_walk.glb. D001 MIGRATION 2026-08-22: this gate rolled its own verdict (a hand `fails` counter + sys_exit) so /api/promote refused it and nothing outside could read its outcome -- nx_gate_green could not judge it and it wrote no harness.jrnl frame, so flake and erosion were invisible for it. Migrated BY HAND onto per-tooth gv_check, NOT via nx_gate_dry_apply (whose candidate collapses N teeth into ONE boolean: ctr[0]=green; ctr[1]=1 -- a one-state judge cannot see a dropped conjunct). T0..T3 keep their exact semantics. `return gv_verdict(...)` so the EXIT CODE carries the verdict -- /api/gate_run derives its verdict from the exit code. WHY THIS GATE WAS REBUILT AT ALL: a fresh build of the unchanged source measured 88,019 B against a served 139,953 B (375 permil, the same shrink as the matched-vintage control). nx_contentdiff named ONE lost run -- `"animations":[{"name":"walk","samplers":[` -- and this gate's own T3 proved on the STAGED binary's ARTIFACT that the walk animation is still declared, so the run was a refactored literal, not lost capability. That proof is what T3 exists for; the promote refusal was about the VERDICT SHAPE, not the content. license_tier: ORIGINAL expect_exit: 0