nx_softtissue.nx

nx_softtissue.nx -- SOVEREIGN soft-tissue solver: XPBD over a tetrahedral physics cage with per-layer compliance. THE tier our body physics did not have. WHY THIS EXISTS (measured, 2026-07-30): the rigmesh viewer drives breasts with a 2-mass EMA spring (a damped oscillator with a calibrated gain). An oscillator has no VOLUME, no LAYERS and no relationship to gravity DIRECTION -- so it physically cannot reproduce the one behavior a human notices instantly: when a woman lies down, soft heavy tissue FLATTENS and spreads toward the chest wall, and firm tissue does so much less. Our own ecosystem already MEASURES that behavior -- nx_breast_drape_pose.nx bands LAYING_BACK at aspect 716..921 / centroid 410..614 against STANDING_FACING 1024..1331 / 614..768 -- so a ruler existed for a simulator that did not. This organ is the simulator; that validator is its oracle (see nx_softtissue_gate.nx). Same solver serves buttocks, thigh and abdomen: the tissue differs only by PROFILE ROWS, never by code (rule 11, rule 25). METHOD (XPBD, Macklin et al. -- position-based, so it cannot blow up like a force solver): per substep: predict -> solve constraints N times -> derive velocity -> damp constraints: distance (edges) | volume (Neo-Hookean incompressibility, C = 6V - 6V0) | contact (sphere/capsule = a finger or a surgical instrument) | attachment (chest-wall particles pinned, inverse mass 0) compliance: alpha_tilde = alpha / dt^2, per LAYER, from data rows. alpha 0 = rigid (muscle, Cooper's ligament), alpha large = soft (adipose). WHY IT CANNOT INVERT: the volume constraint's correction magnitude scales as 1/(6V) as the element collapses, so resistance rises without bound as det F -> 0. Tissue gets HARDER the harder it is pressed -- the same qualitative guarantee the Neo-Hookean strain-energy barrier gives, which is why a deep press dimples instead of exploding. Proven by tooth, not asserted: the gate presses a capsule 60% into the cage and asserts every 6V stays > 0. UNITS: positions in cmm (0.01 mm) -- the NXA VERT unit, so cage and mesh share a frame. H=171530 cmm = 1.7153 m. Scalars q12 (4096=1.0), compliance q20 (1048576=1.0). Velocity cmm/s. dt in microseconds. Frame: +x lateral, +y cranial, +z anterior. Gravity is a UNIT VECTOR in that body frame -- standing (0,-1,0), supine (0,0,-1). That single input is what makes lying-down flattening emergent, not scripted. QUANTIZATION (honest): positions are integer cmm, so one substep's displacement carries ~1 cmm of rounding. st_div_r rounds to NEAREST (not toward zero) so the error does not bias into fake damping -- a truncating divide would silently drain energy and read as viscosity we did not model. license_tier: ORIGINAL No hw writes (Rule 26).