nx_parity_judge_gate.nx

nx_parity_judge_gate.nx -- PROOF the ORGAN EYE classifies like the human did (the operator's bar: "the workflow isnt complete until you [the Nishi team] see the exact same issues i see without your or my involvement"). The truth set = REAL top-20 suite captures a human eyeballed on 2026-07-16: google -> BLANK-NISHI (1) blank render that fooled the parity metric at 97.3% bing -> NO-ORACLE-CONTENT(5) headless Chrome itself served blank wikipedia -> CONTENT-OK (0) the best real render github -> CANVAS-OFF (3) white vs attribute-selector dark theme linkedin -> SPARSE-NISHI (4) full text, missing decoration/columns netflix -> STRIP-COLLAPSE (2) near-black + a hairline strip (the HUMAN said STRIP on 2026-07-16; the pin was held at BLANK only while the render carried <2pm ink. RE-PINNED 2026-08-19 to the human's own word: the 2026-08-18 capture reads 4pm ink and the organ now returns STRIP-COLLAPSE, i.e. it sees what the human saw) ⚠2026-08-19 MEASURED on the laptop's 2026-08-18 capture set (bench/top20 is laptop-resident; this NAS copy abstains): 4 of 6 reproduce the truth set; google's Nishi render is NO LONGER BLANK (search box, buttons, footer render; organ says CONTENT-OK, ink 9pm spread80 368pm) -- the BROWSER improved, so the google pin is a fossil awaiting the operator's eyeball. A seat's read nominates; it does not re-pin. Fixture volatility: paths point at bench/top20 (regenerated by every suite run); if a site's live page changes class legitimately, re-pin WITH an eyeball. MIGRATED onto nx_gate_verdict 2026-08-11 (D001): the old g_case returned 1 for FIXTURE MISSING exactly like a misclassification, so an absent bench/ read as a classifier regression. Missing fixtures are now gv_need preconditions -> verdict=SKIP exit 3 (evidence=none), never RED; the six classification checks are unchanged and their FAIL direction still exits 1 via gv_verdict. license_tier: ORIGINAL expect_exit: 0