nx_worley_noise.nx

nx_worley_noise.nx -- Worley cellular noise (F1 distance) in Q10. Cross-cutting Substrate A primitive per nxc2/docs/NISHI_GAME_ENGINE_ROADMAP.md. Worley 1996 "A Cellular Texture Basis Function" (SIGGRAPH '96). Divide the plane into uniform cells; place one feature point per cell at a hash-determined position; the noise value at any query point is the distance to the nearest feature point (F1). Produces natural-looking cobblestone, rock, dried-mud, water-cell, and lizard- scale textures that Perlin FBM cannot. Algorithm: 1. Cell index (cx, cy) = floor(x / Q), floor(y / Q). 2. Local position within cell (fx, fy) = x - cx*Q, y - cy*Q. 3. For each neighbor cell (cx+dx, cy+dy) with dx, dy in {-1, 0, 1}: a. Hash (ncx, ncy, seed) to a feature-point offset in [0, Q). b. Feature absolute (relative to current cell origin): (dx*Q + ox, dy*Q + oy). c. Squared distance from local query point to feature point. 4. F1 = sqrt(min of those 9 squared distances). 3x3 neighborhood guarantees correctness because the feature point closest to (fx, fy) cannot live more than 1 cell away in any axis when feature points are confined to [0, Q) per cell. Determinism: hash is pure LCG over (cx, cy, seed). Same query + seed -> byte-equal output everywhere. Fixed-point strategy: distances kept in Q20 (Q^2) until the final integer-sqrt converts to Q10. Avoids the precision loss that would hit a per-step Q10 divide. Loss audit: only information-discarding op is the integer sqrt (truncates fractional bit; ~1% error worst case). All other ops exact in i64. genealogy_id: worley_1996_cellular lineage_id: nx_worley_noise_q10_f1