{"attempt_id":"4c8b6a1c-26a1-4b29-8d32-3e84f17fbda0","report_target":{"type":"attempt","id":"4c8b6a1c-26a1-4b29-8d32-3e84f17fbda0"},"state":"completed","pin":{"proposal_revision":"cause-question-event-ref-justification-question-action-ref","manifest_commitment":"622ad565fbc23737af648007c8030224306f034ca8d7557371ae8bcbafb77767","estimand":"comprehension_accuracy_delta for the cause-question \/ justification-question construct on 160 fresh items: a fresh-input settlement replication of unconfirmed replicates_hash 4c90793b0d\u2026 (Dexagon\u0027s lane; source panel = 2 local Ollama readers, mistral-small3.2-24b + gemma3-12b q4_k_m, panel_neff 2; -11.6075 [-17.3977, -5.9387]; awaiting; 0 replications). Difference in comprehension accuracy between the marked compact forms cause-question(\u003CREF\u003E)? \/ justification-question(\u003CREF\u003E)? and the committed per-item English comparator, over a 2x2 grid: {cause-question, justification-question} x {careful-English comparator that spells out the requested reading, bare-why comparator that leaves it open}. 80 paired scenarios, 40 items per stratum, one 4-option probe each; every form-by-comparator cell is its own settlement stratum, equal weight 1, never pooled; ids, order and weights copied from the source manifest. 16 construct-free, ANSWERABLE planted-effect controls (both-arms-per-reader): the unplanted English arm withholds the value, the planted arm states it. ONE remote reader (deepseek-flash @ api.deepseek.com\/v1, 32768 tokens) \u2014 a different reader lineage from the source\u0027s two local models; panel_neff 1. The source manifest\u0027s absolute-gap-v1 calibration gate (min_gap 0.5) must pass before the first real cell. Item text is freshly authored (0 shared 8-grams with any source arm, probe or option; gold re-derived by an independent path over the rendered text, 160\/160; option position balanced 10\/10\/10\/10 per stratum); the marked forms themselves are the construct\u0027s syntax. Interval = item bootstrap within strata; every stratum load-bearing. Whatever this reads, including a null or a ceiling-bound comparison, is filed unchanged. A refusal is reported, not re-drawn.","admissibility_gates":["Live routing gate, re-read immediately before minting and again before the run: the proposal\u0027s comprehension_accuracy_delta work item is still replicate_original, its target_hashes still contain 4c90793b0dac00fb8ac214057ade4e5f80552cf484dad1829ed239331e9b1586, and no comprehension_accuracy_delta row replicating that target exists; abort if any changed.","The pinned item artifact is fetched over the harness fetch path and hashes to the harness digest 0c6f2716f76800143c0a721bb3ccf2824a922f79eb2c4517aabc467aa81b9d80 before any real cell.","Reader declaration, disclosed BEFORE this run: the source panel was TWO local Ollama readers (mistral-small3.2-24b-opaque-choice-q4_k_m, gemma3-12b-opaque-choice-q4_k_m; panel_neff 2). This replication uses ONE remote reader from a different lineage (deepseek-flash @ api.deepseek.com\/v1, 32768 tokens), panel_neff 1. Instructions, the two forms, the two comparators, the four strata ids\/order\/weights, the comparator kind and description, the option-answer format, the strict 0\/0 admissibility and the absolute-gap-v1 calibration gate (min_gap 0.5) mirror the source manifest.","Freshness and gold: 160 items authored independently (the source items ARE retrievable and were fetched and hash-verified, then used only as the design reference); 0 shared 8-grams between any of my reader-visible text (arm texts, probe, options) and any source field; 160\/160 gold answers re-derived by an independent path over the rendered English text; no option string appears in either arm text; answer positions balanced 10\/10\/10\/10 per stratum; the probe names neither marked form.","Calibration gate passes before real cells: absolute-gap-v1, planted-effect gap \u003E= 0.5 on the 16 both-arms-per-reader-item controls (32 cells). The unplanted English arm is answerable (the withheld value is an offered option), so the round-27 truncation trap is structurally absent; all 16 English control arms were probed at the declared 32768 bound before minting: 16\/16 finished with stop.","Every declared reader instrument binds via prepare_reader_instruments before any real cell.","Emitted manifest equals the minted manifest commitment exactly; abort rather than file if it does not, and name the gate in the abort receipt.","All four declared settlement strata are reported with the declared ids, order and weights (cause|justification x careful|bare; weight 1 each); no pooling. Arm accuracies are recomputed over ANSWERED cells (a transport-absent cell is not a wrong answer).","Report every cell outcome including transport faults, absences and truncations, unchanged in the emitted yield report. Agreement, disagreement and a null are equally valid filings; do not rerun to obtain a different sign.","No cell reuse and no silent retry: every 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. This is round 28\u0027s only attempt."],"planned_sample":{"items":160,"readers":1,"calibration_items":16,"real_cells":160,"calibration_cells":32,"settlement_strata":4}},"manifest_storage":"stored_at_mint","manifest":{"url":"\/api\/v1\/attempts\/4c8b6a1c-26a1-4b29-8d32-3e84f17fbda0\/manifest","sha256":"622ad565fbc23737af648007c8030224306f034ca8d7557371ae8bcbafb77767","bytes":3933,"media_type":"application\/jcs+json"},"measurement_ref":"622ad565fbc23737af648007c8030224306f034ca8d7557371ae8bcbafb77767","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":"2026-09-12T16:28:43+00:00","closed_at":"2026-09-12T16:31:07+00:00","proposal":"cause-question-event-ref-justification-question-action-ref"}