nx_voxgrid.nx

nx_tissue.nx -- 3D voxel grid storage with palette packing. Biology naming per [[feedback-naming-discipline-no-industry- competitor-overlap]]: a tissue is a 3D mass of cells with shared structure and function. Composes naturally with existing biology primitives (cells form tissue, tissue holds the 3D structure of an organ). Avoids: - "chunk" (Minecraft trademark proximity, Java HashMap conflation) - "lattice" (already taken by nx_lattice.nx -- 2D Minkowski math) - "voxel_grid" (too descriptive of industry term) - "volume" (generic infra term) THIS IS THE OTHER UNLOCK PRIMITIVE for Minecraft-equivalent on $5 ESP32 class hardware. Pair with nx_palette (already shipped) and a 32x32x32 (32768 voxel) tissue at 3 bits per voxel fits in 12288 bytes. Six tissues arrayed in a 3x2x1 grid form a walkable scene in 72 KiB -- comfortably within ESP32 SRAM (520 KiB total). Composes: nx_palette -- the sub-byte packing layer (nx_palette_pack / nx_palette_unpack); tissue wraps the (x,y,z) -> linear-index addressing on top. nx_budget -- tissue memory consumption is a RAM budget item; caller registers nx_pal_packed_bytes() with budget nx_tier -- bpp scales with tier (3bpp at MCU, 4bpp at family, 8bpp at workstation+ for cache friendliness) nx_pathway -- a tissue lives in a cell; cell budget covers the tissue's packed byte count V1 ships cubic + cuboid tissues with palette indices. Real-time streaming from flash to RAM (chunk paging) is queued; today the caller materializes the whole tissue in RAM. Gap list (V1 honest perf verdict): - no flash-paging (whole tissue in RAM) - no RLE compression atop palette pack (each voxel costs bpp bits) - no greedy-meshing helper (caller's geometry layer) - no neighbor-iteration helper (queued) - no spatial hash (32^3 = 32K voxels still fits direct addressing)