nx_nxa_texm.nx

nx_nxa_texm.nx -- WRITE THE TEXM SECTION: the baked PBR map set carried INSIDE the asset. THE MEASURED GAP THIS CLOSES. nx_asset_floor_gate scores the shipped character 4/7 and prints "PBR texture maps (TEXM) -- we shade procedurally from genome: ABSENT". An independent lane measured that this -- not triangle count -- is why renders read as clay: at 28,092 tris the silhouette is already at 0.94x a shipping real-time budget, and the body carries 104 joints against VRM 1.0's 15 required. Adequate FORM with NO SURFACE is the definition of clay. TEXC (the UV unwrap) has been present in the asset for some time and nx_nxa_texbake produces real maps -- but NOTHING WROTE A MAP BACK INTO THE CONTAINER. Proven absence, not assumed: grep TEXM over 23,018 files, coverage_complete=1 corpus_complete=1, returns exactly TWO hits, both inside nx_asset_floor_gate, both READS. This organ is the missing writer. WHICH PBR MODEL TEXM STORES, AND WHY -- SETTLED HERE RATHER THAN LEFT IMPLICIT. TEXM stores SPECULAR-GLOSSINESS: D(albedo) S(specular) G(gloss) N(normal), exactly the set nx_nxa_texbake_lib bakes (NTB_MAP_ALBEDO/SPEC/GLOSS/NORMAL) and exactly the set knowledge/asset_texture_floor.conf measured from 23 real shipped packages ("tex_map_set 4 maps D(diffuse) S(specular) G(gloss) N|NM(normal)"). Storing anything else would mean transcoding at bake time and recording a number no instrument in this estate produces. !!THE CONSEQUENCE, STATED PLAINLY RATHER THAN DISCOVERED LATER: glTF 2.0 core is METALLIC-ROUGHNESS, and KHR_materials_pbrSpecularGlossiness was ARCHIVED by Khronos. So a TEXM set CANNOT be written to core glTF 2.0 without a conversion, and that conversion is LOSSY -- spec-gloss to metal-rough is not bijective (a dielectric with a coloured specular has no exact metal-rough equivalent). Anyone exporting TEXM through glTF owes that conversion step; this organ does not pretend it is free. The alternative -- baking metal-rough instead -- was NOT chosen because it would make our own measured floor unusable as a comparison. ADDITIVE ONLY (rule 13). Every pre-existing section is copied forward and its checksum is RECOMPUTED rather than copied, because a copied checksum validates its own bug -- the same discipline nx_nxa_texc_lib's ntx_apply already applies. An existing TEXM is REPLACED, not duplicated, so the organ is idempotent: running it twice yields one TEXM, never two. RESOLUTION IS NOT MINE TO PICK. The default is READ from knowledge/asset_texture_floor.conf (tex_res_floor, measured at 4096 px from real package image headers). If the conf cannot be read the organ REFUSES rather than inventing a default -- an unmeasured default is exactly the magic number this estate forbids. A caller may override it explicitly, and the organ ANNOUNCES the bytes each map cost so the resource decision is visible instead of buried. nx_nxa_texm <in.nxa> <out.nxa> [res] [seed] [hue] EXITS: 0 OK | 2 usage | 3 bad container | 4 missing prerequisite section | 1 I/O license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0