nx_nxa_texbake_region_lib.nx

nx_nxa_texbake_region_lib.nx -- PER-REGION SKIN ATLAS + PERIODIC MICRO-RELIEF TILE, written INTO the asset as a TEXR section (2026-08-30, /compare/graphics GR29 gpe_skin_microsurface -- the ASSET half). THE OPERATOR'S BAR, VERBATIM: "for per region we need texture ... make sure we aren't just doing more smooth surface painting." So this organ bakes RELIEF -- a 16-bit height channel in MICRONS and a normal DERIVED from that height by finite differences -- and it MEASURES, per region, what the chart footprint can actually carry before it prints carried=1 for anything. WHAT WAS MEASURED BEFORE A BYTE WAS BAKED. The estate's adequacy ruler (nx_craft_emit cw_atlas_adequate) is span-based: texel = 1.7m / res, 830um at the shipped 2048. But TEXC (nx_nxa_texc_lib) is a per-JOINT cylindrical unwrap: each bone owns one tile of a g x g grid, u runs the FULL circumference, v the bone length. The physical footprint of a texel is therefore (bone circumference / tile px) on u, and a chest bone's circumference is a metre. The `measure` verb below solves the UV->3D Jacobian on every triangle and reports, per region, the area-weighted footprint and the FRACTION OF SURFACE AREA that carries each band -- the honest number. A per-region UNIQUE atlas at 2048 raises the tile px by ~g/gR and carries the primary band on more of the body; it CANNOT carry the 250um secondary band around a torso at any resolution this estate ships. That is arithmetic, and it is why the three finer bands ride a PERIODIC MICRO TILE (nx_relief_lib rlf_relief_um2_tile): a texel pitch that is physical by construction (two per pore orifice), repeated by the shader per chart at (chart extent / tile repeat) using the per-joint circumference and length this section publishes. This is the field's practice for exactly this case: Epic's skin guidance keeps wrinkles in the unique normal and puts pore detail in a tiling texture [@ue-skin], MetaHuman carries a separate Micro Skin Detail normal layer [@mh-materials], the scan vendors ship the micro band as its own displacement channel [@txyz-vface], Disney's procedural micro-wrinkle synthesis places pores first and grows the furrow graph between them [@disney23], and Nagano et al. measured that microstructure is a DISPLACEMENT that smooths under stretch and roughens under compression [@nagano15] -- the next contract this file names (ntr_tension_attenuate) and does not fake. THE AREOLA IS ANATOMY, NOT A DISC. The torso region's albedo and height carry the nipple-areola complex at the MESH-DERIVED bust apex (nx_nxa_dyna's lateral anchors, then the most anterior vertex near each): nipple radius and projection from published anthropometry and the areola from the published 1:3 nipple-to-areola ratio [@nac-review21], the areola darker than the surrounding skin by a lightness step DERIVED from the estate's own referee band against a published breast-skin tone [@nac-tattoo24], Montgomery glands as a discrete dome field. Every one of those numbers is a ROW in knowledge/areola_relief.conf with its source; this organ READS them (nx_lane_conf, the shared getter) and REFUSES to bake anatomy when a row is missing -- an unsourced default is the magic number rule 11 forbids. The registry row nx_anatomy an_add(AF_MONTGOMERY ...) is the vocabulary; the ledger is the provenance. TEXR PAYLOAD (i64 words). ADDITIVE: TEXM (map set 0) is carried byte-for-byte, never touched -- the shipped page reads TEXM's blob at a hard-coded +144, so growing TEXM would break the live sampler.