nx_substrate_layers.nx

nx_substrate_layers.nx -- bits-up layer enum + handoff quality. User directive 2026-05-16: "from the bits to bytes up handoffs so we can get extreme precision on how capable we build out nishi lang". Cardinal feedback-bits-up-canonical-layer-no-reinventing already names the layer stack as an audit discipline ("audit L0(bits) -> L1(containers) -> L2(transforms) -> L3(algorithms)"). This file makes that stack a SEALED-ENUM PRIMITIVE so every substrate file can self-declare its layer + the coverage scanner can produce a bits-up roll-up that shows EVERY layer handoff. The six S-class substrate layers: L0 BITS raw i64 / u8 / pointer arithmetic; no containers L1 CONTAINERS *u8 buffer + length, NxArena, Image, NxTensor, NxStream -- typed shapes over L0 bits L2 TRANSFORMS nx_scale_*, nx_image_*, nx_hash_*, nx_pk_eq_ci -- pure functions over L1 containers L3 ALGORITHMS nx_aes, nx_sha3, nx_ssim, parse, codegen -- stateful or multi-step over L2 transforms L4 COMPOSITES racing primitives, capability audit, gauntlet -- orchestrators over L3 algorithms L5 SURFACES visualizer output, dashboards, CLI wrappers -- user-facing presentation over L4 composites Handoff quality (the precision-of-handoff axis the user named): CLEAN typed, no leakage, sealed-enum verdict DERIVED composed correctly, higher layer omits some typing LEAKY lower-layer details leak through (e.g., raw *i64 where *Image was canonical; the F-meta-4 syscall coupling) BROKEN handoff doesn't compile or has a known regression Per cardinal user-owns-every-bit: substrate carries the SEALED ENUMS (6 layers + 4 handoff verdicts). Which layers a particular user-built substrate slice exercises is caller-driven. nx_capability_claims: