nx_dr_semjudge_cli.nx

nx_dr_semjudge_cli.nx -- judge insights against a document BOTH ways, side by side (DR-10). Shows exactly where the semantic tier beats the lexical one: for every insight it reports the lexical containment score, the PPMI semantic maxsim score, and whether lexical MISSED a match the semantic judge found (lexical 0, semantic > 0). Useful for calibrating the judge threshold and for auditing why a source was or was not credited. nx_dr_semjudge <document_text> <insight_1> [insight_2 ...] e.g. nx_dr_semjudge "the team defeated their rivals" "won" -> {"insights":[{"lexical_permil":0,"semantic_permil":86,"lexical_missed":1}]} ⚠The PPMI model is a count model over QA-benchmark prose; specialist terms are OOV and score 0. Honest tiering: this is not a neural cross-encoder. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0