nx_dr_semjudge_x_cli.nx

nx_dr_semjudge_x_cli.nx -- A/B semantic judge: same code, ANY PPMI model (DR-11). The judge's measured ceiling is out-of-vocabulary terms: a model built over QA-benchmark prose does not know enterprise-research vocabulary, so those terms score 0. To test whether a domain-augmented model lifts that ceiling you must hold the CODE constant and vary only the MODEL -- which a hardcoded model path made impossible. This CLI takes the model as an argument and reports OOV coverage alongside the scores, so the comparison is auditable. nx_dr_semjudge_x <model.bin> <document> <insight_1> [insight_2 ...] Built on the shared nx_ppmi_lib (eats half of debt seq283 -- one loader, not two). license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0