nx_nxa_dyna.nx

nx_nxa_dyna.nx -- WRITE THE DYNA SECTION: per-bone soft-tissue physics DECLARED IN THE ASSET, with the lateral breast-bone pair DERIVED FROM THE MESH'S OWN GEOMETRY. WHY THIS EXISTS. nx_asset_floor_gate measured our shipped character and found 5 midline chest joints and ZERO lateral pair. Real breast bones come in L/R pairs OFF the midline; midline joints at chest height ARE THE SPINE CHAIN, so driving them swings the whole torso -- the wobble the operator kept reporting. With no bone to drive, chest motion had to be faked at the shader vertex band, and the engine's own joint picker (|bind.x| < 9000, i.e. near-midline) was selecting spine joints as stand-ins using a superseded height normalisation (z/171530, which ignores the mesh origin -- the same error that once put the nipples below the legs). THE FIX IS A FORMAT CAPABILITY, NOT ANOTHER SHADER PATCH. FBX and its class carry no physics at all: a rig travels as geometry plus a joint tree, and every consumer re-invents the dynamics in engine code. DYNA makes the ASSET carry its own soft-tissue contract -- anchor, axis, stiffness, damping, travel clamp, influence radius, falloff -- so the engine DRIVES DECLARED DATA instead of hardcoding bands. That is the difference between a mannequin and a body. ANCHORS ARE MEASURED, NEVER GUESSED. For each side we take the CENTROID of the front-hemisphere vertices inside the bust band. Anatomy therefore comes from the mesh in front of us, so a different body type yields different anchors -- and the firm/bouncy RANGE the operator asked for falls out of geometry rather than a table of magic numbers. Physics constants are NOT invented here: they are the banked, in-band values already proven by the chest-motion oracle (2.5 Hz, 3.2 cm travel, 0.0 cm torso, 0.89 coherence) and by nx_softbind (chest K base 60 step 25, C base 100 step 16). Moving them out of code and into the asset is the point -- rule 11, and the recombinator wants them as data. nx_nxa_dyna <in.nxa> <out.nxa> ADDITIVE AND IDEMPOTENT: the input is never modified; an existing DYNA is REPLACED, not doubled, so running twice is identical to running once. DYNA payload (i64 words): [0]=n_bones [1]=stride(12), then per bone: 0 side(-1 L,+1 R,0 mid) 1 anchor_x 2 anchor_y 3 anchor_z 4 k_q8 5 c_q8 6 max_disp 7 influence_radius 8 falloff_q8 9 axis_mask(1=x,2=y,4=z) 10 k_step_q8 11 c_step_q8 license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0