{"$defs":{"configuration_id":{"pattern":"^gpqa_cfg_v1_[0-9a-f]{64}$","type":"string"},"delta":{"additionalProperties":false,"properties":{"denominator":{"minimum":1,"type":"integer"},"numerator":{"type":"integer"},"percentage_points":{"pattern":"^(?:\\+|-)?(?:0|[1-9][0-9]?|100)\\.[0-9]{2} pp$","type":"string"}},"required":["numerator","denominator","percentage_points"],"type":"object"},"evaluation_run_id":{"pattern":"^gpqa_run_v1_[0-9a-f]{64}$","type":"string"},"evidence_variant_id":{"pattern":"^gpqa_evd_v1_[0-9a-f]{64}$","type":"string"},"release_id":{"pattern":"^gpqa-clean-results-v[0-9]+\\.[0-9]+\\.[0-9]+$","type":"string"},"row":{"additionalProperties":false,"properties":{"attempts_per_question":{"minimum":1,"type":"integer"},"clean":{"$ref":"#/$defs/score"},"clean_question_n":{"const":189},"configuration_date":{"format":"date","type":"string"},"configuration_id":{"$ref":"#/$defs/configuration_id"},"delta":{"$ref":"#/$defs/delta"},"delta_direction":{"enum":["increase","decrease","unchanged"]},"evaluation_run_id":{"$ref":"#/$defs/evaluation_run_id"},"evidence_class":{"enum":["project_native","direct_strict","direct_compatibility","targeted_recovery","derived_uniform"]},"evidence_variant_id":{"$ref":"#/$defs/evidence_variant_id"},"model_configuration":{"minLength":1,"type":"string"},"observed_competition_rank":{"minimum":1,"type":"integer"},"organization":{"minLength":1,"type":"string"},"original":{"$ref":"#/$defs/score"},"original_question_n":{"const":198},"protocol":{"pattern":"^mean@[1-9][0-9]*$","type":"string"},"release_id":{"$ref":"#/$defs/release_id"},"schema_version":{"const":"gpqa-results-public-leaderboard-v2"},"validation_profile":{"minLength":1,"type":"string"}},"required":["schema_version","configuration_id","evaluation_run_id","evidence_variant_id","model_configuration","organization","protocol","attempts_per_question","evidence_class","validation_profile","original","clean","delta","delta_direction","observed_competition_rank","original_question_n","clean_question_n","release_id","configuration_date"],"type":"object"},"score":{"additionalProperties":false,"properties":{"denominator":{"minimum":1,"type":"integer"},"numerator":{"minimum":0,"type":"integer"},"percent":{"pattern":"^(?:0|[1-9][0-9]?|100)\\.[0-9]{2}%$","type":"string"}},"required":["numerator","denominator","percent"],"type":"object"}},"$id":"https://gpqa.ai/schemas/gpqa-clean-results-leaderboard-v2.json","$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"data_through":{"format":"date","type":"string"},"ranking":{"additionalProperties":false,"properties":{"aggregation":{"const":"combined_mean_at_n"},"compute_normalized":{"const":false},"metric":{"minLength":1,"type":"string"},"ordering":{"minLength":1,"type":"string"},"protocol_interpretation":{"minLength":1,"type":"string"},"tie_policy":{"const":"competition exact-rational ties"}},"required":["aggregation","metric","ordering","tie_policy","protocol_interpretation","compute_normalized"],"type":"object"},"release_id":{"$ref":"#/$defs/release_id"},"rows":{"items":{"$ref":"#/$defs/row"},"minItems":1,"type":"array"},"schema_version":{"const":"gpqa-clean-results-leaderboard-v2"},"summary":{"additionalProperties":false,"properties":{"configuration_count":{"minimum":1,"type":"integer"},"evidence_class_counts":{"additionalProperties":false,"properties":{"derived_uniform":{"minimum":1,"type":"integer"},"direct_compatibility":{"minimum":1,"type":"integer"},"direct_strict":{"minimum":1,"type":"integer"},"project_native":{"minimum":1,"type":"integer"},"targeted_recovery":{"minimum":1,"type":"integer"}},"required":["project_native","direct_strict","direct_compatibility","targeted_recovery","derived_uniform"],"type":"object"},"protocol_counts":{"additionalProperties":false,"properties":{"mean@1":{"minimum":1,"type":"integer"},"mean@16":{"minimum":1,"type":"integer"},"mean@2":{"minimum":1,"type":"integer"},"mean@6":{"minimum":1,"type":"integer"},"mean@7":{"minimum":1,"type":"integer"},"mean@8":{"minimum":1,"type":"integer"}},"required":["mean@1","mean@2","mean@6","mean@7","mean@8","mean@16"],"type":"object"}},"required":["configuration_count","protocol_counts","evidence_class_counts"],"type":"object"}},"required":["schema_version","release_id","data_through","ranking","summary","rows"],"title":"GPQA Diamond Clean public leaderboard","type":"object"}
