nx_id_palette.nx

nx_id_palette.nx -- THE CORRELATED IDENTITY PALETTE. /compare/procgen C33, symbol gn_palette_sample. The first consumer of nx_id_param_core, and the direct answer on the GENERATOR side to the operator's "our colour palette is barely used". WHAT WAS THERE BEFORE, MEASURED: nx_breeding carries exactly ONE colour trait, GN_T_HUE, drawn as an independent uniform. nx_skin_ita then maps that single scalar onto a straight L* locus at fixed a* and fixed b*. So a cast of N characters occupies a ONE-DIMENSIONAL, UNCORRELATED colour space: nothing about a character's hair relates to anything about her skin, because there is no hair gene at all -- the live page tints the iris with the same scalar. WHAT THIS EMITS: three swatch genes -- skin, hair, eye -- drawn from ONE shared melanin factor with a per-tissue loading, so they COVARY. Each lands on nx_breeding's own 0..1000 range, so a swatch is a drop-in for the single hue trait it supersedes rather than a parallel colour system. THE LOADINGS ARE THE MODEL'S ONLY FREE PARAMETERS AND THEIR ORDERING IS THE FALSIFIABLE CLAIM: skin is the most strongly determined by the shared melanin factor, hair next, eye least -- iris colour is the most independently varying of the three. That ordering, not the exact values, is what the gate tests, and the values live in knowledge/id_palette.conf with their provenance. THE INCUMBENT IS KEPT RUNNABLE below as the control. A baseline you cannot re-run is a number and not a control, so the gate measures the old behaviour at runtime instead of quoting a figure. LIB, no main (ecosystem convention). license_tier: ORIGINAL No hw writes (Rule 26).