nx_dialogue_gate.nx

nx_dialogue_gate.nx -- REFEREE for WRITING rung W-DLG-1 (nx_dialogue). [T1] turn + session counting and side identification. [T2] per-turn intensity scoring is exact (benign proxy lexicon, per the seam). [T3] all four cells, one session of each kind -> exact cell id. [T4] THE HEADLINE METRIC on a 4-session corpus: tally = [1,1,1,1], unprompted rate = 250 permil, and the PARTITION SUMS to the session count (a partition is a claim: check the parts). [T5] first-explicit-turn index + NEG-CONTROL (a clean session returns -1, not 0). [T6] escalation slope: rising positive, flat zero, falling NEGATIVE (sign discriminates). [T7] NEG-CONTROL the detector can FAIL TO FIRE: raise the threshold above every score and the same corpus collapses to 4 NEITHER -- proving the meter is not a constant that always accuses. [T8] whole-word: a longer word CONTAINING a lexicon term does not score (heated != overheated). [T9] division guard: rate over an empty corpus is 0, not a trap. Evidence -> stdout + knowledge/status/dialogue_gate.log. Exit 0 GREEN / 1 RED. Sovereign x86_64. license_tier: ORIGINAL expect_exit: 0