nx_galx_audit.nx

nx_galx_audit.nx -- audit the ENTIRE cid->path store. Every entry. No sampling. Operator 2026-08-07: *"not to do it only partially but have full ecosystem scope as you seem to do sample size bullshit"*. Fair, and specific: after joining 109 gen images I reported "5/5 serve 200" and called the job verified. Five. Of a hundred and nine. Of two hundred and thirty-five thousand. ★★★★★★ A SPOT CHECK PROVES THE MECHANISM CAN WORK, NEVER THAT THE POPULATION DOES. The whole class of defect this session has been about -- 0 of 109 joined, a stale index, a collision that silently ate an image -- is INVISIBLE to a sample drawn from the head of a file, because the head is the oldest and most-tested part of any append-only store. ★★★★★ SAMPLE THE HEAD OF AN APPEND-ONLY FILE AND YOU HAVE SAMPLED ITS PAST, NOT ITS PRESENT. So: walk every bucket, resolve every path, stat every file. The cost is one stat per entry and it is worth paying, because the answer "how much of this gallery is actually backed by bytes" has apparently never been asked. Usage: nx_galx_audit <idx> <blob> [--list-missing N] license_tier: ORIGINAL