_cdc_fixture_gen.nx

_cdc_fixture_gen.nx -- LOCAL INSTRUMENT (2026-09-06): a seeded pseudo-random fixture generator and a chunk-length census for the content-defined chunker, so a dedupe accept rule is measured on ENTROPY-BEARING bytes. WHY: the first 64 MiB dedupe measurement used a generated PNG whose IDAT is a stored deflate block of a period-3 byte pattern (1 MiB gzips to 5,678 B). A gear fingerprint over period-3 content takes three values, none hit the mask, so EVERY cut was forced at max (1339 x 48,402 + 28,970 = 64,839,248 exactly) -- fixed-size chunking wearing a content-defined name -- and a one-byte insertion shifted every later cut, so reuse stopped at the insertion point (661 of 1340 = the offset-preserving prefix, 49.3 percent = 32,000,000 / 64,839,248). That is a property of the DATA every chunker shares; the accept rule needs a fixture whose cuts are content-defined. verbs: gen <out> <bytes> <seed> xorshift64* stream, deterministic for (bytes, seed) insert <in> <out> <offset> <byte> copy with ONE byte inserted at offset (the dedupe workload) chunks <file> <min> <avg> <max> cut with nx_cdc_lib at the door's own parameters; prints count, min/max/avg length, how many cuts were FORCED at max, and the gzip-free entropy proxy (distinct bytes in the first block) exit: 0 ok | 2 usage | 1 io