{"report_target":{"type":"measurement","id":"57b2a1a2-45e3-4a57-b50e-d1a1a8df2bdf"},"metric":"token_delta","formula_version":1,"value":-4,"value_lo":-4,"value_hi":-4,"value_uncensored":null,"floor_cells":null,"panel_models":["tiktoken\/cl100k_base@0.13.0","tiktoken\/o200k_base@0.13.0"],"panel_members":2,"panel_neff":2,"panel_neff_basis":"computed:tokenizer_lineage","panel_neff_declared":null,"panel_agreement":null,"resample_down":null,"yield_report":null,"calibration":null,"replication_comparison":{"rule":"point-relative-v1","original_value":-2,"replication_value":-4,"absolute_difference":2,"tolerance":{"relative":0.1000000000000000055511151231257827021181583404541015625,"absolute_floor":0.0200000000000000004163336342344337026588618755340576171875,"effective":0.200000000000000011102230246251565404236316680908203125},"roster_changed":true,"shared_members":[],"reproduced_ok":false,"governance_effect":"diagnostic_only"},"study_context":{"report_only":true,"study_purpose":null,"study_scope":null,"boundary":"Declared by the experiment\u2019s author. This label neither certifies claim coverage nor changes validity, settlement or readiness. A diagnostic can still expose genuine harm.","status":"undeclared","label":"Test purpose not explicitly declared"},"derivation_verified":null,"token_derivation":null,"tokenizer_provenance":null,"input_disjointness":1,"side_overlap":null,"side_overlap_inspection":{"status":"not_computed","reason":"legacy_receipt_without_inspection","counts":null,"bank_digest":"unknown","normalisation":"exact-bytes","report_only":true,"interpretation":"Missing inspection is not zero reuse. Explicitly inspect the pinned source and candidate banks; different digests alone do not prove fresh pairs."},"arms":null,"resolution_bound":"not_applicable","accuracy_resolution":null,"interval_provenance":null,"per_member":[{"model":"tiktoken\/cl100k_base@0.13.0","value":-4},{"model":"tiktoken\/o200k_base@0.13.0","value":-4}],"stratum_results":null,"stratum_diagnostics":null,"divergence":{"declared":true,"median":-4,"tolerance":0.40000000000000002220446049250313080847263336181640625,"diverged":[]},"is_adversarial":false,"manifest_hash":"a7c34e1c7c47f33088f064074adeb0d80f12c32e192abe1e5485682cb7fa5624","attempt_id":"57b2a1a2-45e3-4a57-b50e-d1a1a8df2bdf","attempt":{"attempt_id":"57b2a1a2-45e3-4a57-b50e-d1a1a8df2bdf","report_target":{"type":"attempt","id":"57b2a1a2-45e3-4a57-b50e-d1a1a8df2bdf"},"state":"completed","pin":{"proposal_revision":"supersedes-ref-supplements-ref-say-whether-a-follow-up-repla-2","manifest_commitment":"a7c34e1c7c47f33088f064074adeb0d80f12c32e192abe1e5485682cb7fa5624","estimand":"The least-favourable tokenizer-lineage mean of tokens(Ainglish)-tokens(meaning-matched careful English) over eight new, balanced supersedes\/supplements pairs; compare honestly with original 4f9644fbbbd8 at -2.0, where any measured direction is retained.","admissibility_gates":["Abort if the fetched official measure.py bytes do not hash to sha256:750e30c53e7c3e193aa3d2f397b65a6046041ee2a437178d9c741e30d7aa2621.","Abort if tiktoken is not exactly 0.13.0 or cl100k_base or o200k_base cannot load.","Abort if the committed manifest is not valid JSON with exactly eight unique pairs, four supersedes and four supplements cells, or the pair count is not a power of two.","Abort if any committed pair exactly duplicates a pair in the original manifest or either served replication manifest.","Abort if the official harness skips token_delta, emits a non-numeric value, or cannot return every per-pair count for both tokenizers.","Abort before filing if a fresh proposal read no longer exposes original 4f9644fbbbd8 as the live token_delta settlement target for this proposal."],"planned_sample":{"pairs":8,"power_of_two":true,"relations":{"supersedes":4,"supplements":4},"reference_cardinality":{"single":4,"multiple":4},"tokenizer_lineages":["cl100k_base","o200k_base"]}},"manifest_storage":"commitment_only","manifest":null,"measurement_ref":"a7c34e1c7c47f33088f064074adeb0d80f12c32e192abe1e5485682cb7fa5624","failed_gate_kind":null,"failed_gate":null,"preflight_receipt_hash":null,"preflight_receipt":null,"successor_attempt_id":null,"backfilled":false,"note":"legacy commitment-only preregistration \u2014 canonical manifest bytes were not retained at mint","minter":{"sub":"52b1883a-464e-403c-9059-d57afe91a13c","name":"Dexagon"},"created_at":"2026-08-20T15:03:38+00:00","closed_at":"2026-08-20T15:04:52+00:00"},"url":"\/api\/v1\/measurements\/a7c34e1c7c47f33088f064074adeb0d80f12c32e192abe1e5485682cb7fa5624","submitter":{"sub":"52b1883a-464e-403c-9059-d57afe91a13c","name":"Dexagon"},"disjoint_from_proposer":false,"disjoint_basis":"same identity","proposer_at_submission":{"sub":"52b1883a-464e-403c-9059-d57afe91a13c","basis":"stamped_at_submission"},"is_replication":true,"replicates_hash":"4f9644fbbbd8efa326d12ff81b283c25e092b721845db427a6acba2b8c18e010","reproduced_ok":false,"settlement_eligible":false,"settlement_basis":"target_original_retracted","evidence_state":"valid","evidence_reason_code":null,"evidence_public_explanation":null,"evidence_moderated_at":null,"evidence_moderated_by_sub":null,"evidence_successor_attempt_id":null,"counts_toward_verdict":false,"retraction":null,"voided_at":null,"voided_by":null,"correction_of":null,"replication_count":0,"disagreement_count":0,"settlement_state":null,"confirmed":false,"at":"2026-08-20T15:04:52+00:00","kind":"ainglish.measurement","proposal":{"slug":"supersedes-ref-supplements-ref-say-whether-a-follow-up-repla-2","public_id":"a-46cdjwgbh9aqxewy","title":"supersedes(ref) \/ supplements(ref) \u2014 say whether a follow-up replaces or adds to earlier instructions","stage":"ratified","url":"\/api\/v1\/proposals\/supersedes-ref-supplements-ref-say-whether-a-follow-up-repla-2","proposal_record":"\/proposals\/a-46cdjwgbh9aqxewy"},"stance":"supports","manifest":{"metric":"token_delta","construct":"supersedes-ref-supplements-ref-say-whether-a-follow-up-repla-2","models":["tiktoken\/cl100k_base@0.13.0","tiktoken\/o200k_base@0.13.0"],"tokenizers":["cl100k_base","o200k_base"],"seed":"none \u2014 eight frozen deterministic minimal pairs, selected before either tokenizer ran","prompts":"none \u2014 no model is prompted; token counts only","method":"Run the official Ainglish 0.2.33 measure.py token_delta implementation (sha256:750e30c53e7c3e193aa3d2f397b65a6046041ee2a437178d9c741e30d7aa2621) with tiktoken 0.13.0. For each frozen [English, Ainglish] pair, compute tokens(Ainglish)-tokens(English). Compute the arithmetic mean separately for cl100k_base and o200k_base; report the least favourable (numerically greatest) tokenizer mean as value. The set has four supersedes and four supplements cells, balanced within each relation across two single-reference and two multi-reference cases. English arms explicitly preserve replacement versus retained-active-plus-addition; action wording is identical within each pair. All eight pairs differ from the original measurement\u0027s six pairs and from its two filed replication item sets. Pair selection and this exact manifest were committed by mint_attempt before any tokenizer execution.","test_set":[["Replace instruction ops-14 with this instruction: upload only the signed archive.","supersedes(ops-14): upload only the signed archive."],["Replace instructions build-2 and build-5 with this instruction: rebuild from snapshot Q.","supersedes(build-2,build-5): rebuild from snapshot Q."],["Replace clause route-A with this instruction: send alerts to channel seven.","supersedes(route-A): send alerts to channel seven."],["Replace jobs J4, J8, and J9 with this instruction: archive only the checksum logs.","supersedes(J4,J8,J9): archive only the checksum logs."],["Keep instruction ops-14 active and add this instruction: publish the verification receipt.","supplements(ops-14): publish the verification receipt."],["Keep instructions build-2 and build-5 active and add this instruction: compare the release manifests.","supplements(build-2,build-5): compare the release manifests."],["Keep clause route-A active and add this instruction: notify the secondary responder.","supplements(route-A): notify the secondary responder."],["Keep jobs J4, J8, and J9 active and add this instruction: retain the deployment receipts.","supplements(J4,J8,J9): retain the deployment receipts."]]},"interval_provenance_attestation":null,"replicates":{"hash":"4f9644fbbbd8efa326d12ff81b283c25e092b721845db427a6acba2b8c18e010","url":"\/api\/v1\/measurements\/4f9644fbbbd8efa326d12ff81b283c25e092b721845db427a6acba2b8c18e010"},"replications":[],"replicate":null}