nx_fetch_any_gate.nx

nx_fetch_any_gate.nx -- teeth for the intelligent fetch dispatcher (operator 2026-07-14: researcher/census/ adversary/critic must "not get errors and return nothing without informing the user"). ADVERSARY DISCIPLINE: the classifier is LAW only if it discriminates a REAL page from the exact thing that fools a naive fetcher -- a Cloudflare 200-OK CHALLENGE page. Gates fa_status / fa_challenge / fa_is_real / fa_body_off / fa_to_body. license_tier: ORIGINAL D001 MIGRATION 2026-09-01 (/compare/mediaingest). This gate had no elf, and the reason was not neglect: /api/promote REFUSES a gate that rolls its own verdict, because nothing outside can read its outcome -- nx_gate_green cannot judge it and it records no harness.jrnl frame, so flake and erosion stay invisible. The teeth here were already good (13 of them, with a real adversary case), which is precisely why the refusal was expensive: a well-designed gate that cannot be promoted proves nothing at all. EVERY SUBJECT, ORDER AND ASSERTION IS UNCHANGED. Only the verdict machinery moved onto nx_gate_verdict: gv_ctr/gv_check make declared == executed by construction (the hand-rolled pass/tot incremented both from the same line, so a tooth that stopped running lowered nothing and still read GREEN), and gv_verdict puts the outcome in the exit code where the judge can reach it. The four teeth that assert a REFUSAL are renamed neg-control-* so the L2 census can SEE them -- they always were negative controls, and were invisible purely by naming.