nx_body_gen_winding_candidate_t365.nx

nx_body_gen.nx -- ★THE PRODUCT: a SOVEREIGN PROCEDURAL human body, generated from the first byte up. No scanned mesh, no third-party asset: the body is EMITTED from anthropometric parameters (fractions of stature, per-mille). BodyParts3D is the ORACLE we measure against -- never the product. Competitor platforms are benchmarked on CAPABILITY only; none of their assets are used. v2 (GX-23): the body is now built from CONTINUOUS RING PROFILES, not disjoint cylinders. Each part (torso, arm, leg) is ONE tube threaded through control rings and sub-sampled, so there are no seams where segments used to butt together -- that was the largest silhouette defect vs the oracle. Hands and feet are ring extensions of the arm/leg tubes, so they attach continuously. Normals include the taper slope, so shading follows the true surface rather than a stack of cylinders. nx_body_gen <out.nxmesh> <height> [radial] [sub] [relief] [canonpath] [fat] [prof] [?] [hf] ★★THE CANON PATH IS argv[6] AND IT WAS UNDOCUMENTED (seq1377). It is the single most important input -- this emitter carries NO geometry literals, so WHICH canon you hand it is the whole product -- and its absence from this line is why the nx_skullgen -> mesh chain LOOKED non-composable and cost a measurement cycle. Defaults: relief/RLF argv[5], canon argv[6] (knowledge/canon_male.dat), fat argv[7], prof argv[8], hf argv[10]. ★A usage line that omits a positional does not merely under-document it -- it makes every LATER positional unreachable, because you cannot skip one. license_tier: ORIGINAL expect_exit: 0