nx_nxa_morf.nx

nx_nxa_morf.nx -- WRITE THE FACE AND MORF SECTIONS: the addressable face region and the expression morph basis, both DERIVED FROM THE ASSET'S OWN GEOMETRY and carried inside it. THE MEASURED GAP THIS CLOSES. nx_asset_floor_gate scores the shipped character 5/7 and prints face geometry (FACE) -- we PAINT the face from ratios instead: ABSENT -- gap to the floor blend shapes / morph targets (MORF) -- no expressions without these: ABSENT -- gap to the floor Those two report lines ARE the spec: the gate asks whether a section with that tag is present. But the same gate carries the law that makes a bare tag worthless: PRESENCE OF A SECTION IS NOT PRESENCE OF THE CAPABILITY. So this organ never writes an empty section. FACE carries the real vertex membership of the head region and MORF carries a per-vertex displacement for every channel, with an evaluation verb that proves the deltas actually move geometry. WHY THE DELTAS ARE DERIVED AND NOT IMPORTED. Full-population measurement of every authored source the estate can reach: 4 of 4 Fab FBX donors: 0 BlendShape, 0 BlendShapeChannel deformers 3 of 3 glTF donors: 0 targets, 0 targetNames, 0 weights seed-san.vrm: DOES carry targetNames (VRoid authors expression morphs) The VRM morphs are indexed against the VRM's OWN topology. This character is our procedural nishi being with a different vertex count and a different order, so copying those deltas onto it would be FABRICATION wearing a measurement's clothes. The honest source for OUR mesh is the estate's OWN parametric basis: nx_faceanat already derives a face from the anthropometric canons and exposes faceanat_morph_width / _length / _jaw / _eyes / _lips / _nose / _cheek. Those SEVEN channels are the target set here -- taken from our own canon work, not invented names, and not another project's vertex deltas. 2026-08-23 RUNG 2 -- THE CHANNEL SET IS NOW DATA. The first version hardcoded SEVEN channels from nx_faceanat's canon morphs. Seven against a best-of-breed MetaHuman LOD0 (669 blendshapes AND joints) is roughly 95x, the largest single gap on the board. The channel table now lives in knowledge/facs_au.conf and is ADOPTED from Apple ARKit's published ARFaceAnchor.BlendShapeLocation vocabulary -- MEASURED from a hash-pinned mirror, not quoted from memory. Adding an action is a data edit; this organ's logic never names a channel. NO PICKED NUMBERS. The head band is nx_headcrop's measured band (it OWNS that number; this organ cites it). Every sub-band boundary is a facial canon nx_faceanat already encodes: the RULE OF THIRDS (hairline-brow-subnasale-chin, equal) and the RULE OF FIFTHS (face width is five eye-widths, the nose occupying the central fifth). Every anchor is a centroid MEASURED from the asset. The stature axis is DERIVED as the largest extent, never assumed. license_tier: ORIGINAL No hw writes (Rule 26).