nx_id.nx

nx_id.nx -- monotonic IDs and 128-bit content addresses. Three ID surfaces, each with explicit guarantees: nx_seqid_next(ctx) monotonic 64-bit, unique per ctx since construction. Use for spans, requests, tasks -- any ordering you only need within one process lifetime. nx_uuid7(ctx) time-ordered 128-bit ID, RFC 9562 layout draft (UUIDv7). Two i64 outputs. The 48 high bits are unix-ms timestamp, remaining 80 bits are random+counter. Use across process boundaries when you need both ordering AND collision-resist. nx_xid_blake_short 8-byte digest of an arbitrary byte buffer using a tiny mixing function (xxhash-style). Use for content addressing where full SHA-256 is overkill. NOT cryptographic. For crypto use sha256.nx / sha512.nx. Why this file exists separately: ID generation is touched by every observability module (nx_log seq, nx_trace span_id, nx_metrics labels) and every IPC layer (request_id headers). Centralising the policy prevents drift -- e.g. one module producing 64-bit decimal, another 128-bit hex, another timestamp-prefixed. All flow through these three primitives. Design references: RFC 9562 (UUIDv7) -- time-ordered v7 layout Twitter Snowflake -- 64-bit time+machine+seq id ULID spec -- Crockford-base32 26-char id xxHash3 -- non-crypto fast mixing function