{"id":"W2732894280","doi":"10.1371/journal.pbio.2002212","title":"Can cancer researchers accurately judge whether preclinical reports will reproduce?","year":2017,"lang":"en","type":"article","venue":"PLoS Biology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University; McGill University","funders":"Canadian Institutes of Health Research","keywords":"Replication (statistics); Pace; Optimism; Biology; Cancer; Cognitive psychology; Psychology; Statistics; Computer science; Data science; Social psychology; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"category_scores_codex":[0.592562,0.0009382469,0.001728906,0.0058661,0.00166521,0.008458762,0.0033036,0.004713345,0.00292827],"category_scores_gemma":[0.9046748,0.001286645,0.002196406,0.003704297,0.007822348,0.01289352,0.004387408,0.004077838,0.002539392],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003256039,"about_ca_system_score_gemma":0.005117523,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003136801,"about_ca_topic_score_gemma":0.00237315,"domain_scores_codex":[0.5114905,0.3509485,0.05372865,0.02589617,0.05477761,0.003158666],"domain_scores_gemma":[0.04252473,0.7082071,0.095927,0.1068116,0.04432778,0.002201837],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004406281,0.000347519,0.4906427,0.003849558,0.00393604,0.0005386407,0.03642913,0.01040076,0.004075301,0.02656324,0.01976994,0.3990409],"study_design_scores_gemma":[0.001361994,0.004384408,0.6024257,0.007878451,0.002818376,0.003245712,0.01154593,0.04781826,0.02420183,0.1865305,0.1063704,0.001418464],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5184523,0.02277436,0.3288544,0.06866322,0.004220347,0.004869807,0.004524424,0.002025516,0.04561547],"genre_scores_gemma":[0.923067,0.001609827,0.06599162,0.005086141,0.001190326,0.001219859,0.000718948,0.0001973655,0.0009189724],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.407438,"threshold_uncertainty_score":0.5024437,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8966954757372453,"score_gpt":0.6814503299149918,"score_spread":0.2152451458222535,"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."}}