{"artifacts":[{"bytes":717,"path":"CHANGELOG.md","sha256":"a77b106b46aa3966694aa08549587985dcaf60a7e2eba3f03d41744403628013"},{"bytes":888,"path":"CITATION.cff","sha256":"496bc8a59acdc326780da3b37f7dc7a976dbb037a5bfe6817942b1cda3f238a6"},{"bytes":1420,"path":"LICENSES.md","sha256":"2448da9b7e25affe9af68d6d99bae885881ee3b52c8b09a61caddd1fc6d72e21"},{"bytes":3341,"path":"README.md","sha256":"a8e718ce6a3782b04b2dd7e02a79d32706811b67c65d6fe34e72b817972ba542"},{"bytes":3469,"path":"category-summary.json","sha256":"e3278a70d3bb4f27d746d4523061405325eb4487366161a78b4407cd327a0ecf"},{"bytes":660345,"path":"configuration-manifests.json","sha256":"46d1a1ac0099f09d74ba86ee49dbdf35a3cbf5f4970e52ab0507dd2f68b2c3ea"},{"bytes":5717,"path":"coverage-gaps.json","sha256":"7860495476676fc525c0f30d269a052a4e445c213c26e65b5c92c8edc148fe77"},{"bytes":100613,"path":"leaderboard.csv","sha256":"a9532e89f66d0a7d9bd30eeb4c6ea2af0339c3e2e9680ede5ea954476336cd6b"},{"bytes":252328,"path":"leaderboard.json","sha256":"df13958c9226c727ddc435b320740d699de5a25b5d5681d34417c49d20be83e0"},{"bytes":4305,"path":"methodology.json","sha256":"f35bbebc0a5105b0a70059485e8cbde0d2aa4c58f73a829c3419aa2a96c9e07d"},{"bytes":117586,"path":"score-changes.json","sha256":"cf1840cf3bc3f56eb67bbb2c9ae7749c66ae69aa02cfe73793d8be7e69923c0f"},{"bytes":101007,"path":"uncertainty.json","sha256":"56699a6d9ecde8ed5a6c5d4d96d656cee4539c7245053d28427e31791126e239"},{"bytes":1813,"path":"schemas/category-summary.schema.json","sha256":"4c39481af73b03204d5aeac1bd40cea67a5d0f8d9ec4e60199afca6433409f20"},{"bytes":44112,"path":"schemas/configuration-manifests.schema.json","sha256":"b7a840ceda84bbf0c2085a76da9b40e2a3e1938970be5b5fc22fb23df3a15e7d"},{"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":31,"ranked_configurations":265,"source_artifact_rejections":15,"source_ledger_rows":295,"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":"2026-09-25T01:00:00Z","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":"2026-09-25","id":"gpqa-clean-results-v0.1.10","public":true,"publication_status":"published","timestamp_utc":"2026-09-25T01:00:00Z","title":"GPQA Diamond Clean Results","version":"0.1.10"},"repository":{"base_commit":"07f21bbb8b48ab25abd710a545558b7ed4eaa76e","exporter_source_sha256":"27f413d3316379859ecfd7e27affe7d943f37d60bc4f7c15ae27d4c670acbefa","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":"7f5a90d77d282b41051db51e6f44c215688333a84675bb315043d6c87274ccb2"},"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":"878f2c1ac27281343d232ae2eb6c843e0a9b059c4d1ccd216f5e126fc472989b","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":"3.14.7","pyyaml":"6.0.3"},"uncertainty":{"confidence_level":"95%","method":"synchronized_record_id_cluster_bootstrap","replicates":20000,"seed":20260817}}