nx_world_quality_grader.nx

nx_world_quality_grader.nx -- mathematical verdict for procgen worlds. Per user 2026-05-16: "build an iterative loop like we got going on elder ai i dont want to have to manually check if you generate good maps. does minecraft or these other prc gens have something that actually mathematically measures if the world is good like how god says and i saw it was good as a confirmation check its not trash" Genesis 1:31 -- "And God saw every thing that he had made, and, behold, it was very good." This primitive ships the substrate's SABBATH GRADER for generated worlds: a per-tick verdict over a heightmap that returns sealed-enum grade + per-axis Q14 scores + refine-axis directive so the caller can iterate. Most procgens (Minecraft, Houdini, No Man's Sky, Spelunky) have NO mathematical world-quality measurement -- generation is one-shot noise with hand-tuned constants. The substrate's posture is honest-perf-verdict: every output has a measurable verdict, and every LOSE carries a named_improvement. This primitive surfaces that posture at the WORLD level. FULL CAPABILITY (per feedback-maximum-capability-no-simplification cardinal, 2026-05-16): 8 axes: DIVERSITY -- (max-min) / max_possible_relief. Range-based. LOCAL_VARIATION -- mean |neighbour-difference| / max_relief. Banded (too jagged is also bad). EXTREMES -- ratio of cells in top-10% + bottom-10%. Signature features push this up. FLATNESS_PENALTY -- ratio of cells at modal-bucket height. HIGHER = MORE FLAT = WORSE; verdict inverts. FRACTAL_DIM -- Hurst-exponent estimator via multi-scale mean-abs-diff slope. Spehar 2003: most aesthetically pleasing fractal dimension band maps to H in [0.3, 0.7]. Peak band 0.4-0.6. BIOME_COHERENCE -- fraction of cells where biome ID matches expected elevation band (DESERT at low, FOREST at mid, SNOW at high). Caller-supplied biome_map. Skipped (verdict=MARGINAL) if NULL.