nx_night_grade.nx

nx_night_grade.nx -- SOVEREIGN night-reading benchmark gate (Reader arc, rung W-NIGHT-1). MEASURES the sourced eye-fatigue mechanism (blue light) per reader theme. The palette SSOT now lives in the NATIVE content-addressed seg_store (ncfg / nx_native_config) -- NOT a flat .tsv (operator 2026-06-20: "stop using tsv, nishi ecosystem only, all rungs everywhere"). The gate SEEDS the 6-theme palette into a fresh per-run store then reads it back BY COLUMN NAME, so it has ZERO dependency on knowledge/registry/night_themes.tsv (migrated off + removed). For each theme it computes, from the dominant emitted colors (bg + fg, sRGB 0..255): blue_emit = bb*BG_W + fb*FG_W (emitted blue, area-weighted) total_emit = (br+bg+bb)*BG_W + (fr+fg+fb)*FG_W blue_frac = blue_emit*1000 / total_emit (per-mille -- "how blue is the light", coverage-independent) BG_W:FG_W = 9:1 = documented assumption that page background covers ~90% of the screen. Sourced rationale (knowledge/research/2026-06-16-night-reading-sourced.md): short-wavelength (blue) light drives the circadian/eye-fatigue response (melanopsin/ipRGC); warm 2700-3000K cuts blue; sRGB display white = 6500K (bluish). The night win = LESS BLUE, by composition -- not merely darkness. Negative controls prove the metric measures blue, not dark. KAT order in the store (positional rows): 0 light, 1 sepia, 2 dark, 3 night, 4 black, 5 coolnight(neg). expect_exit: 0 (all assertions green) license_tier: ORIGINAL