nx_dr_semjudge_gate.nx

nx_dr_semjudge_gate.nx -- KAT + neg-control for the SEMANTIC judge tier (DR-10). Runs BOTH judges on the same input to prove the escalation is real: a source that says "defeated" scores ZERO under the lexical judge for an insight saying "won", but scores above zero under the PPMI semantic judge. That is the measured lift, on the real shipped 41MB model -- not a fixture. Neg-controls: an out-of-vocabulary insight scores exactly 0 (no spurious match), and the known-measured synonym pairs must outrank the unrelated ones. Runs from the nxc2 CWD so knowledge/index/semppmi_v1.bin resolves. DRY nx_gate_verdict lib.