nx_tier.nx

nx_tier.nx -- substrate-wide tier configuration. Single point of edit for scale-agnostic substrate. Per user directive 2026-05-13: "with the i64 it looks hardcoded everywhere if we really want this dynamic dont we want that to be a changeable value everywhere so it can switch to i128 and i256 etc." Per cardinals: - feedback-numeric-tier-ladder.md (N0..N9 swap) - feedback-scale-agnostic-substrate.md (MCU..HPC swap) - feedback-substrate-additive-not-restrictive.md (declare cost) SEMANTIC ALIASES (not all should swap simultaneously): nx_int -- DEFAULT ARITHMETIC integer. Swappable across the numeric tier ladder. Swap this to i128 to make the entire substrate compute in 128-bit integers. nx_size -- MEMORY-SIZE integer. Always platform-pointer-width. Used for buffer sizes, mmap byte counts, struct sizes. Does NOT swap with nx_int -- changing this would break pointer arithmetic. Stays i64 on RV64. nx_idx -- ARRAY-INDEX integer. Same width as nx_size on flat-memory targets. Distinct alias so future GPU/distributed targets can change indexing without touching arithmetic. nx_byte -- The byte type. Stays u8. Distinct alias so MCU targets that emulate u16-byte memory could rebind. HARDWARE-TIER BUFFER SIZES (declare cost, don't restrict): NX_BUF_TINY -- 64 B (MCU-friendly; stack-safe) NX_BUF_SMALL -- 256 B (MCU heap-friendly) NX_BUF_MEDIUM -- 4096 B (page-size; workstation default) NX_BUF_LARGE -- 64 KiB (server-friendly) NX_BUF_HUGE -- 1 MiB (HPC; assumes virtual memory) Use these instead of `sys_mmap(4096)` etc. so the substrate