nx_campaign_verify.nx

nx_campaign_verify.nx -- DID THAT CAMPAIGN ACTUALLY DO WHAT IT CLAIMED, FOR EVERY ROW? WHY IT EXISTS, and it is a defect I committed today rather than one I read about. On 2026-08-07 I ran two convergence campaigns (139 files, then 228) and "verified" them like this: - byte identity: checked ALL rows -> honest - import resolution: checked THE FIRST 40 OF 228, reported "ALL IMPORTS RESOLVE" - surplus-line clusters: read the TOP 22 of 1,119 distinct lines, called it the distribution Every one of those numbers was true about the sample and asserted about the population. The operator named it exactly: "sample size bullshit ... full ecosystem scope". => ★★★★★★A VERIFICATION THAT DOES NOT PRINT ITS DENOMINATOR IS A SAMPLE WEARING A PROOF'S CLOTHES. => ★★★★★★A CHECK I HAVE TO REMEMBER TO RUN IN FULL IS A CHECK THAT WILL BE RUN IN PART. The estate already banked why: A CAPABILITY OFFERED AS ADVICE IS ADOPTED AT ADVICE RATES; THE SAME CAPABILITY PLACED IN THE PATH IS ADOPTED AT 100%. So this is an ORGAN the campaign CALLS, not a discipline the operator maintains. WHAT IT PROVES, over the WHOLE adopted list, never a prefix: 1. EVERY adopted row is byte-identical to the source it was adopted from (the adopt really landed) 2. EVERY import edge in EVERY adopted file resolves in the TARGET tree (the tree stays coherent) and it prints checked/total for both, so a shortfall is visible rather than inferable. !!IT REFUSES AN EMPTY OR UNREADABLE LIST. A gate that reports "0 of 0 failed, GREEN" on a list it could not open is the absence-tooth defect the estate has already paid for twice: A TOOTH THAT ASSERTS AN ABSENCE PASSES LOUDEST WHEN NOTHING RAN AT ALL. rows==0 is RED, not GREEN. !!IT IS A VERIFIER, NOT A GATE, AND THE NAME NOW SAYS SO. /api/ship REFUSED the first version under D001: "this gate rolls its own verdict instead of inheriting nx_gate_verdict, so nothing can read its outcome -- nx_gate_green cannot judge it and it records no harness.jrnl frame, so flake and erosion stay invisible for it." There is an allow_own_verdict=yes escape hatch and taking it would have been exempting my own gate from the convention it exists to enforce. The deeper error was the NAME: gates here are no-arg and sweep-run, and a thing that takes a population in argv is a tool. => ★★★★★★THE CONVENTION-ENFORCING MECHANISM CAUGHT THE CONVENTION-ENFORCING TOOL. A GUARD YOU EXEMPT YOURSELF FROM IS NOT A GUARD, IT IS A PREFERENCE. => ★★★★★NAMING A TOOL `*_gate` MAKES A CLAIM ABOUT HOW IT IS RUN AND JUDGED -- MEET THE CLAIM OR CHANGE THE NAME. Verdict + journalling are inherited from nx_gate_verdict (gv_check / gv_verdict), so nx_gate_green can judge this like anything else and each run leaves a harness.jrnl frame. Exit: gv_verdict's rc (0 GREEN, 1 RED) | 2 usage | 3 IO. DIALECT: plain-if, <=6 params, consts above use, mmap'd static POINTERS (a BSS static array crashes the module at startup -- banked gotcha). license_tier: ORIGINAL expect_exit: 0 No hw writes (Rule 26).