nx_nxa_pect.nx

nx_nxa_pect.nx -- GIVE THE SOFT TISSUE ITS BONES: append the lateral pectoral pair to an NXA and re-bind the bust vertices to it. WHY THIS EXISTS. The reference-figure rig carries lPectoral/rPectoral as first-class bones (measured 2026-08-03 from morph formula targets across three independent packages); our rig has 104 joints and ZERO off-midline at chest height, which is the root cause behind the vertex-band soft-tissue workaround (debt 1785798366). The DYNA section already declares MEASURED lateral anchors (nx_nxa_dyna derives them from the mesh's own bust-band centroids); this organ promotes those declared anchors into actual SKEL bones with skinning weights, so the engine can drive a bone instead of faking one at the vertices. ★THE TRAP THIS DESIGN AVOIDS (found by reading the consumer BEFORE building): the page runtime has NO joint hierarchy -- every joint poses from its OWN absolute ANIM track, and a joint with no track renders the IDENTITY transform. Bones appended naively would leave a FROZEN CHEST PATCH on an animating body. So each new bone receives a byte-copy of its donor chest joint's track(s) with the header jointIdx rewritten: copied (R, dt) applied at the new bone's OWN bind position is exactly rigid attachment under the consumer's dual-quat T = b + dt - D*b. FAIL-CLOSED MEASUREMENTS, NEVER ASSUMPTIONS: - the SKIN weight layout (planar [i,i,i,i,w,w,w,w] vs interleaved [i,w,i,w..]) and the weight basis (the constant per-vertex sum) are DETECTED from the data and REFUSED if ambiguous; - every ANIM track must be chn==2 (the one layout the consumer parses) or we refuse; - a rig that already carries an off-midline bust-band joint is REFUSED (idempotence: run twice and the second run tells you it is done rather than doubling bones); - all checksums are RECOMPUTED, never copied (nx_nxa_dyna's law: a copied checksum would validate its own copy bug). nx_nxa_pect <in.nxa> <out.nxa> (input never modified) license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0