nx_meet_trust_gate.nx

nx_meet_trust_gate.nx -- liar-kill, MEASURED gate for R6 (the anti-fake engine). Runs the scorer over a labeled fixture set (4 AI-slop resumes + 4 real, 4 ghost postings + 4 real) and proves a MEASURED exceed over the status-quo "accept-all" baseline (incumbents largely DON'T screen for AI-slop/ghost = accept everything). Teeth: (a) our accuracy must be near-perfect AND catch every fake, (b) the accept-all baseline must do far worse, AND (c) the scorer must NOT trivially flag everything (real samples must pass) -- so neither trivial extreme passes. expect_exit: 0