{"id":"W3003585144","doi":"10.1038/s41562-019-0813-1","title":"The Confidence Database","year":2020,"lang":"en","type":"article","venue":"Nature Human Behaviour","topic":"Research Data Management Practices","field":"Computer Science","cited_by":161,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"Economic and Social Research Council; National Institute of Mental Health; Medical Research Council; Agence Nationale de la Recherche; Wellcome Trust; U.S. Department of Health and Human Services","keywords":"Confidence interval; Computer science; Database; Information retrieval; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01567242,0.001278004,0.001962881,0.008433528,0.001173775,0.006299169,0.005147673,0.00324886,0.04796677],"category_scores_gemma":[0.1467427,0.001232744,0.001861607,0.009215537,0.0008469066,0.004132661,0.004977684,0.002993889,0.03612367],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001512267,"about_ca_system_score_gemma":0.004310099,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003847629,"about_ca_topic_score_gemma":0.005458747,"domain_scores_codex":[0.9867346,0.003479608,0.003724393,0.002625823,0.002991739,0.0004438343],"domain_scores_gemma":[0.901556,0.0539225,0.008296284,0.02272517,0.01136043,0.002139554],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001290934,0.0002750424,0.01860669,0.007147967,0.0006716876,0.0001964423,0.0006757588,0.002963865,0.001024229,0.0157759,0.8482288,0.1031426],"study_design_scores_gemma":[0.0005458009,0.0001701797,0.01720808,0.001689686,0.0003125817,0.0003486172,0.0002872063,0.003925942,0.001676636,0.03160179,0.9420596,0.0001740349],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.003056166,0.001199943,0.01827605,0.0007834345,0.0002202404,0.0007524813,0.9616998,0.005788924,0.008223025],"genre_scores_gemma":[0.01550679,0.0007889832,0.02672784,0.0008140855,0.00009835619,0.005368365,0.9470323,0.001300325,0.002363009],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.9843276,"threshold_uncertainty_score":0.1604649,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09731803816968139,"score_gpt":0.3883518749160733,"score_spread":0.2910338367463919,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}