{"artifacts":[{"bytes":716,"path":"CHANGELOG.md","sha256":"3d3ee08d9e6a9d7c590a2351ce185cb93a0048f131f04d73d3e380cbf5f6d1a4"},{"bytes":887,"path":"CITATION.cff","sha256":"067571a725dba16ae6b93b363408ee239626315ee808733ad7653e166ff7d3e7"},{"bytes":1419,"path":"LICENSES.md","sha256":"8d877963d5d5a8e81471e25a6f8b70553b66cb053fd569018e95ff85a21898fc"},{"bytes":3340,"path":"README.md","sha256":"13ae678098f4193621833ee71057e6e1847321af322590b39c522324c69c918b"},{"bytes":3472,"path":"category-summary.json","sha256":"47885dbcf63ae9f7d789e9543c3ee48d8c55af11c91c8fee8ac3018ad8f3504c"},{"bytes":597160,"path":"configuration-manifests.json","sha256":"fbc42209daefde3157352a6f35e0708162c28117d2cf522207ac26b062e4ed8b"},{"bytes":5716,"path":"coverage-gaps.json","sha256":"e0d13ea339aa06b17d03bcea78e544c2eb835aacea807c8ea110ea4d64c5b3a7"},{"bytes":92133,"path":"leaderboard.csv","sha256":"e131ef33761da7bf083a02bb4d52a8226c3a0e6c37b6993c161cc13f2723b52d"},{"bytes":231041,"path":"leaderboard.json","sha256":"5d4c4f02b69cfbc9c5393dd1d4ee049667d427e44ed98881be0b7b2f539486fb"},{"bytes":4298,"path":"methodology.json","sha256":"6eda59d2e27ab3ac545092d9102171c77922d09048ee8b2b80842dfd06d084a4"},{"bytes":107991,"path":"score-changes.json","sha256":"0e6b127e6c8a043e3d9786a1d9aa60ea2a8fc46571f17580f3fc492a721b23f8"},{"bytes":92668,"path":"uncertainty.json","sha256":"9a37d47c92766c078547bfa10716c923d8888550ed0432fc71232b06adbda0f6"},{"bytes":1813,"path":"schemas/category-summary.schema.json","sha256":"4c39481af73b03204d5aeac1bd40cea67a5d0f8d9ec4e60199afca6433409f20"},{"bytes":12758,"path":"schemas/configuration-manifests.schema.json","sha256":"94a5227759026c9712c2e33dbfb56fbdfd18c2223dd4acbfa6c409d0e45f40a6"},{"bytes":1435,"path":"schemas/coverage-gaps.schema.json","sha256":"fa141ea846ea29e00a9615b422b79a046805d9e5b244683472a883a924d1b712"},{"bytes":4393,"path":"schemas/leaderboard.schema.json","sha256":"6cdfe7692e70fabae9e9f8a9580403bd0384757d95775387f46a0e3e57551543"},{"bytes":6961,"path":"schemas/methodology.schema.json","sha256":"63ab84769372f9c85ffd8dc251e219d91912a2a607fb512dad8a18e52096b790"},{"bytes":7420,"path":"schemas/release-manifest.schema.json","sha256":"a814944ad378b68e668cb79702bcf2c3023475de1eac8c9d28fa23de5c061968"},{"bytes":3896,"path":"schemas/score-changes.schema.json","sha256":"27b238c82bdcf811e1816e12ec50cb6ccb66f6fa310b3e16ee94a4d28e321f3b"},{"bytes":2705,"path":"schemas/uncertainty.schema.json","sha256":"5c7cec3cedcd01694fffe9c90c1425d746ab8b46ba79161872637cc796a3bac3"}],"benchmark":{"excluded_question_count":9,"exclusions_sha256":"d51f84fe1df413b062b3ce1b8c7f93ecd1f29484711e9019bdd1d019c25292a5","manifest_sha256":"65edff43aa69d39a7e97fe6fc2a78433c0c77a62af7fd0be8bac20833e770a96","name":"gpqa-diamond-clean","original_question_count":198,"repository_commit":"3c0478bb2854a7ce8e8b4606c3717182f03bfeea","retained_question_count":189,"version":"1.2.2"},"bundle_contract_version":"gpqa-clean-results-public-bundle-v2","citations":[{"license":"CC BY 4.0","title":"GPQA: A Graduate-Level Google-Proof Q&A Benchmark","url":"https://arxiv.org/abs/2311.12022"},{"license":"CC BY 4.0 for data and documentation","title":"GPQA-Diamond-Clean","url":"https://github.com/adamallcock/gpqa-diamond-clean"},{"license":"See source repository","title":"When the Answer Key Is Wrong","url":"https://github.com/adamallcock/answer-key-audit"},{"license":"Source attribution required","title":"Epoch AI GPQA Diamond evaluation evidence","url":"https://epoch.ai/data-insights/gpqa-diamond"}],"contracts":{"exporter":"gpqa-clean-results-exporter-v2","identity":"gpqa-results-identity-v1","leaderboard_row":"gpqa-results-public-leaderboard-v2","privacy":"gpqa-clean-results-answer-safe-public-v1","ranking":"gpqa-results-ranking-v2","release_config":"gpqa-results-release-config-v2","uncertainty":"gpqa-clean-uncertainty-contract-v2"},"counts":{"epoch_derived_uniform_configurations":6,"epoch_direct_configurations":224,"epoch_targeted_recovery_configurations":4,"independent_native_configurations":9,"ranked_configurations":243,"source_artifact_rejections":15,"source_ledger_rows":273,"superseded_source_rows":4,"top_50_coverage_gaps":12,"unresolved_diagnostics":0},"disclosure":{"minimum_configurations":5,"minimum_questions":5,"publish_configuration_by_question_matrix":false,"publish_record_ids":false,"question_output":"aggregate_only"},"independence":"Independent research project; not affiliated with or endorsed by benchmark authors, evidence sources, or model providers.","integrity_note":"The manifest inventories every payload except itself and checksums.sha256. checksums.sha256 covers the manifest and every inventoried payload; it necessarily excludes itself.","public_boundary":"Aggregate answer-safe public release only. No GPQA task content, Record IDs, model-by-question matrix, raw evidence, completions, or reasoning.","ranking":{"aggregation":"combined_mean_at_n","cohort":"all_fully_accepted_configurations","metric":"clean_accuracy","normalize_inference_compute":false,"tie_policy":"competition_exact_rational"},"release":{"canonical_selection_cutoff_utc":"2026-08-17T00:00:00Z","canonical_selection_timestamp_policy":"All hash-pinned accepted inputs are assigned the frozen release timestamp only for deterministic canonical selection. This administrative timestamp is not evidence creation, run, first-seen, or real admission time.","data_through":"2026-08-17","id":"gpqa-clean-results-v0.1.0","public":true,"publication_status":"published","timestamp_utc":"2026-08-17T00:00:00Z","title":"GPQA Diamond Clean Results","version":"0.1.0"},"repository":{"base_commit":"28104a8700fa3b5fd7a98325a99c1ec88536e341","exporter_source_sha256":"bd54d9e3e3a3e9364fa1a4020df3a5314c8422ef8bcf8760a102538e67c5b310","note":"Published public release. The bundle contains aggregate answer-safe results only; private evidence remains outside the public boundary.","uncertainty_builder_source_sha256":"17de1a015e6c0a19235267723b00da43c7b0edf2740188ca56bd36c353c1f3e0","uncertainty_derivation_sha256":"5f823133f846fb9ebdd2309c03fe05c4175d8fb9fa297b9e003e2bc49f9e880c"},"schema_version":"gpqa-clean-results-release-manifest-v2","source_inputs":{"answer_safe_epoch_import_sha256":"14fb13becc9ccf6038d623f14bc5bc192c74d8935db6a1b75956fb73e8f55e6b","epoch_csv_snapshot_sha256":"df4e96c72d8675620122d28d791ff973313c4334019eb1e8046316f8ca196904","epoch_source_artifact_bindings_sha256":"6452ef57bdfc3bf0cade2798a3c1672f08d4ce96c6d1e84e2c1ecaf466921896","independent_native_import_sha256":"3f9c0f1351df8a38c9f5d89eaadce950106865f72788218f178bc9a1d2e5b02a","question_taxonomy_source_sha256":"7bc6bfa4d648beb3a1ab76521f04a3be7311a423691b443c071d85b4afd5ec8b","targeted_recovery_composite_sha256":["827adc26ebf257f27db79f78c90b71f48f9a5d04d4b3ec42853bab642e9d0905","8dcb6b950148d43aded364bcbac178e6281fc41ac0624e1eedd61be36d489887","7c7e19d6362979db9b4a77969a2498f836a9149b9ed71ab40a5d128cbbfa8e6e","6fe0936192f274bf7196d7121679631d893ddde0ee3520941dcfe4e61b4f5e3b"],"uniform_recovery_audit_sha256":"307f478ab700ed681d764619287f471b1878b6568a3cbc5035477dce15c70424"},"toolchain":{"dependency_lock_sha256":"495e8a55e4db01affde20716d21c6d5067f9970ff9e3babb3d7536a959471398","inspect_ai":"0.3.240","jsonschema":"4.26.0","numpy":"2.5.0","python":"3.14.6","pyyaml":"6.0.3"},"uncertainty":{"confidence_level":"95%","method":"synchronized_record_id_cluster_bootstrap","replicates":20000,"seed":20260817}}