nx_gltf2mesh_rig_gate.nx

nx_gltf2mesh_rig_gate.nx -- THE GATE FOR R22, RIG-CARRYING glTF INGEST (contract g2_skin_weights). SUBJECT: the nx_gltf2mesh ELF, forked for real, asked for an .nxa output. The claim under test is that a rigged glTF donor's SKELETON AND SKIN SURVIVE INGEST -- the exact thing nx_roundtrip measured as lost on 2026-08-22 (9 named parts -> 1, 192 bytes = 8 x 24 dropped layer records, on every rigged donor). So the load-bearing teeth read the emitted NXANIM01 back through INDEPENDENT readers, not through the subject's own status line: - nx_asset_floor_gate (a shipped, separately-written NXA reader) reports verts/tris/joints; - nxa_find from nx_nxa.nx (the ONE checksum definition every NXA reader imports) verifies the TOC and every section's payload checksum; - this gate's own minimal TOC walk checks the SKIN/SKEL word lengths against what the donor's vertex and joint counts REQUIRE (computed, never eyeballed) and sums every vertex's weights. FIXTURES ARE THE REAL DONORS, BY ABSOLUTE PATH, PLUS ONE FIXTURE ASSEMBLED AT RUNTIME: the count-mismatch donor is nishi_being_walk.glb with ONE accessor's count patched (8180 -> 8189, same text length so the glb stays well-formed). A hand-authored glb would test the fixture's authoring; a patched real donor tests the subject. Outputs are UNLINKED AT SETUP -- a gate that is not idempotent reports on its first run and lies about every run after (measured on this estate today). DERIVED EXPECTATIONS, QUOTED FROM THE DONORS' OWN JSON (read 2026-08-22, offset 20 of each glb): nishi_being_walk.glb: POSITION count 8180, indices count 49080 (-> 16360 index triples), skins [{"joints":[1..11]}], node 2 "Lsh" translation [-185,520,15] under node 1 "root" (no translation) -> joint 1 binds at (-185, 520, 15) mm and joint 0 is the root (parent -1). ⚠JOINT COUNT IS NO LONGER QUOTED FROM THIS DONOR -- see RG_WALK_JOINTS. Quoting a published artifact freezes ITS vintage as the spec, and on 2026-08-26 the donor was republished with the spec-correct FLOAT normals AND the emitter's current 11-joint rig, turning two teeth RED for a reason that was never about the subject. The other expectations here (vertex/index counts, the Lsh bind) are STRUCTURAL and unchanged by that republish; only the rig grew. nishi-in-seedsan-proportions.glb: POSITION 6728, indices 40332 (-> 13444), joints [1..11], node 2 "Lsh" translation [-65,520,15]. synth/elf.glb: skins ABSENT (skin_joints=0) -- the unskinned control. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0