nx_ppp_probe_gate.nx

nx_ppp_probe_gate.nx -- THE VERDICT nx_ppp_probe COULD NOT PUBLISH (2026-08-25). nx_ppp_probe scores PERSON / PRODUCT / PLACE on our own live surface AND on each rival's front door, and those numbers are PUBLISHED as competitive claims on /compare. It shipped with a 20-tooth internal --kat and NO external gate, so /api/gate_run could not read its verdict and no roster ever ran it: a published metric whose health nobody could ask about from outside. WHY IT NEEDED ONE, measured 2026-08-25: the probe scored viewport=0 on /games, a page that CORRECTLY emits its mobile-scaling declaration in SINGLE quotes. It matched the double-quoted and the bare forms and not the single-quoted one, so a PUBLISHED place score was understated by 50 permil. The defect carried its own control in the same measurement -- on that identical page `canonical` read 1, because THAT check had been hand-written with all three quote forms. A FIX APPLIED IN ONE PLACE AND NOT ITS SIBLINGS IS HALF A FIX, AND THE HALF LEFT UNDONE IS THE ONE THAT SHIPS. SCOPE, DECLARED: this gate FORKS the deployed elf, so it is END-TO-END. It proves the subject's own self-test still RUNS, has not SHRUNK, and still DISCRIMINATES on live pages. It does NOT re-derive the scoring -- the subject's --kat owns that, and duplicating it here would be a second ruler. SUBSTITUTABLE: argv[1] overrides the subject, so nx_gate_bite can hand it a freshly built mutant. license_tier: ORIGINAL No hw writes (Rule 26).