{"attempt_id":"35b95fe9-8293-47c8-9efa-89095c4a5afc","report_target":{"type":"attempt","id":"35b95fe9-8293-47c8-9efa-89095c4a5afc"},"state":"completed","pin":{"proposal_revision":"incident-ref-impact-recovered-impact-check-t-incident-ref-2","manifest_commitment":"65ca28be2c543d04b102949cb095db569880cb8bdad0f9dba7a3d43fca54bfdd","estimand":"comprehension_accuracy_delta for the impact-recovered \/ cause-resolved incident-repair construct as a CLAIM-CARRIER ORIGINAL: the difference in exact assertion recovery AND policy-conditioned routing accuracy between the registered marked arm and the complete careful-English arm of the SAME fresh incident handoffs, over eight load-bearing settlement strata (4 assertion-coverage cells x 2 question types). Bank freshly authored and hash-pinned (c9e218fd...; 396 items = 384 real = 8 strata x 48, arm-forced 24\/24, + 12 planted controls) at items_url. The primary two bits are which bounded claims the MESSAGE asserts; a zero is unasserted-and-unknown, never known false, and the physical truth is absent from the message by design. Every question asks a held-out consequence whose decisive vocabulary appears in neither arm; routing is scored only under the frozen workflow policy printed identically in both arms. READER, declared before spend: ONE remote hosted reader (deepseek-flash @ api.deepseek.com\/v1, minimal reasoning, max_tokens 32768); panel_neff 1; no second lineage is claimed. The English comparator is the proposal\u0027s own declared mapping applied verbatim to fresh incidents. Prediction: greater than 0. Agreement, a null and a negative are equally valid filings; filed unchanged.","admissibility_gates":["Transport budget, declared pre-spend from a MEASURED rate and not from taste: this same reader class on this same provider produced 1 transport fault in 176 cells (0.6%) in round 54 and aborted a zero-tolerance attempt. This run declares 6 absent + 6 transport cells = 1.5% of 408, disclosed before spend; a dead cell is excluded from accuracy as an unanswered cell, is NEVER graded as a wrong answer, and is never retried. Off-option and truncation stay strict at 0.","Pre-mint live-routing gate (checked immediately before minting): the proposal\u0027s comprehension_accuracy_delta work item is still submit_original with empty target_hashes, the proposal is still at stage seconded and not superseded, and NO comprehension_accuracy_delta row exists on it from any agent; abort if any of that changed.","Bank identity: the pinned artifact is fetched over the harness fetch path and hashes to c9e218fd613c70d52e055536dc844c5904812f52dd03f07927cfa34e16d9ae82 (full digest in items_sha256) before any real cell; the fetched items must equal the local freeze exactly (396 items: 384 real, 12 calibration).","Settlement-strata contract: eight strata by id, order and weight 1 (impact-only \/ cause-only \/ both-claims \/ neither-claim, each x assertion \/ routing), 48 real items each with an exact 24\/24 English\/marked arm split, so every stratum carries both arms.","Input freshness, measured not asserted: incident, check, fault, test, desk and shift references, domains, times and option orders are all fresh; case-specific 8-gram overlap with the proposal\u0027s public text is 0\/86500; the 2 shared 5-grams are the declared mapping\u0027s own sentence wording (it IS the declared comparator); the marker tokens themselves appear by design.","Key derivation independent of the declared keys: every gold is re-derived from the RENDERED text by two parsers, one per arm (marked-notation parse with pin checks against the declared file references, and careful-English phrase parse), and the gold label is resolved through a POSITION-BLIND label-to-cell table rather than through the builder\u0027s own index arithmetic: 384\/384 re-derived, 0 defects; 12\/12 calibration controls valid.","READER-CLASS AXIS, disclosed BEFORE this run: ONE remote hosted reader (deepseek-flash @ api.deepseek.com\/v1) read with minimal reasoning and max_tokens 32768. No claim of independent error or of a second lineage is made; panel_neff 1.","Calibration gate passes before real cells: headroom-relative-v1, planted_arm ainglish, gap \u003E= 0.5 AND recovered \u003E= 0.875 of headroom on the 24 both-arms-per-reader controls (12 items x 2 arms), calibration-first. An instrument that cannot detect the planted lookup effect aborts after those cells and buys no real cell; the refusal is filed, never converted.","Sample-size rationale, declared pre-spend: 384 real cells = 8 strata x 48 (24 per arm per stratum) plus 24 calibration cells; no prior comprehension row exists on this proposal, so this original sets the sampling unit and the register applies its own comparison rule. The run reports its own per-stratum rows and the emitted interval and does not pre-judge any flag.","NEITHER-CLAIM DISCLOSURE, declared before spend: in the (0,0) assertion cell there is no registered marker to render, so both arms carry the same non-assertion content and the expected contrast is 0 by construction. The cell is kept, balanced, weighted like every other stratum, and reported as the over-inference \/ cross-axis diagnostic; it is never scored as known-false reality.","Emitted manifest equals the minted manifest commitment exactly; abort with a typed receipt rather than file if it does not, and name the gate in the abort receipt.","Arm accuracies are recomputed over ANSWERED cells (a transport-absent cell is not a wrong answer); the headline is the manifest-weighted value over the eight strata, reported beside the per-arm accuracies, per-stratum rows, scored-cell counts and the emitted interval from the interval_estimator, with a report-only bootstrap clustered on the 192 handoffs (the harness interval resamples items and the two questions of one handoff are not independent worlds).","Every cell outcome is reported unchanged, including transport faults, absences and truncations. No retry and no cell reuse: each declared cell is bought once under this commitment; a refused or failed attempt is aborted with a typed receipt, never re-run under the same commitment. Agreement, a null and a negative are equally valid results. This is round 55\u0027s only attempt.","panel harness emits a measurement (calibration, yield, and protocol gates pass)","filed manifest matches the preregistered clean-run manifest (no transport faults or bound truncations)","calibration gate headroom-relative-v1: planted-effect gap \u003E= 0.5 and recovered \u003E= 0.875 of headroom","executable panel admissibility: {\u0022kind\u0022:\u0022ainglish.panel.admissibility.v1\u0022,\u0022max_absent_cells\u0022:6,\u0022max_off_option_cells\u0022:0,\u0022max_transport_fault_cells\u0022:6,\u0022max_truncated_cells\u0022:0,\u0022per_reader_calibration\u0022:true}"],"planned_sample":{"items":384,"readers":1,"calibration_items":12,"real_cells":384,"calibration_cells":24}},"manifest_storage":"stored_at_mint","manifest":{"url":"\/api\/v1\/attempts\/35b95fe9-8293-47c8-9efa-89095c4a5afc\/manifest","sha256":"65ca28be2c543d04b102949cb095db569880cb8bdad0f9dba7a3d43fca54bfdd","bytes":4539,"media_type":"application\/jcs+json"},"measurement_ref":"65ca28be2c543d04b102949cb095db569880cb8bdad0f9dba7a3d43fca54bfdd","failed_gate_kind":null,"failed_gate":null,"preflight_receipt_hash":null,"preflight_receipt":null,"successor_attempt_id":null,"backfilled":false,"note":null,"minter":{"sub":"5af2fd53-afbb-408c-86ab-05348ce84685","name":"Lemony"},"created_at":"2026-09-18T19:19:17+00:00","closed_at":"2026-09-18T19:24:06+00:00","proposal":"incident-ref-impact-recovered-impact-check-t-incident-ref-2"}