nx_crummy_detector.nx

nx_crummy_detector.nx -- direct "is this world crummy?" detector. Per user 2026-05-16: "i just dont want them to be crummy" Crummy worlds share a small set of failure modes that statistical graders (nx_world_quality_grader) sometimes miss because they reward heightmap diversity even when the OVERALL experience is bland. This primitive grades the 7 failure modes that produce "crummy" output regardless of underlying statistics: AXIS 0 -- RANGE_USAGE: Does the heightmap use the available relief budget? Worlds with all hills at the same height (small range relative to the body's max relief) are crummy. AXIS 1 -- LOCAL_CONTRAST: Is neighbour-to-neighbour height variation perceivable? Smooth-noise worlds score high on diversity but low here -- they look like a putting green. AXIS 2 -- SIGNATURE_PEAK_COUNT: Does the world have AT LEAST a few extremes (peaks above 0.7 of relief; valleys below 0.1)? Uniform mid-elevation worlds score 0 here. AXIS 3 -- BIOME_DIVERSITY: If a biome map is supplied, is biome distribution non-degenerate? An all-FOREST or all-DESERT world is crummy. Shannon entropy of the biome histogram. AXIS 4 -- PALETTE_RICHNESS: If a color palette is supplied, are there at least 5 distinct hues? Monochrome worlds are crummy. AXIS 5 -- HORIZON_VARIATION: If horizon ray samples are supplied, does the silhouette have meaningful stddev? Pancake-horizon worlds are crummy. AXIS 6 -- SPATIAL_NON_REPETITION: Auto-correlation check. Does the heightmap repeat at scale N? Worlds that tile (because the noise was sampled with too-low frequency) are crummy. EMITS LAYER_VERDICT (16 i64) with kind = NX_LAYER_KIND_READABILITY (the closest existing kind; could split to a CRUMMY kind later). CRUMMY-IS-NOT-A-LOSS CONVENTION: each axis is HIGHER = LESS CRUMMY. Caller composes through nx_meta_verdict for the God-said-good check. Skipped axes (null inputs) score MARGINAL (= 0.5Q) so they don't drag the verdict. Caller can supply biome_map = 0, palette = 0, horizon = 0 to skip those axes.