{"id":"W3130133331","doi":"10.4300/jgme-d-20-00111.1","title":"A Systematic Review of the Relationship Between In-Training Examination Scores and Specialty Board Examination Scores","year":2020,"lang":"en","type":"review","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"Smiths Detection (Canada)","funders":"","keywords":"Specialty; Medical education; Objective structured clinical examination; Medicine; MEDLINE; Educational measurement; Family medicine; Psychology; Medical physics; Political science","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008328901,0.0003539104,0.002816195,0.0009359472,0.00009557211,0.00003133917,0.0004473439,0.0004179781,0.0000462269],"category_scores_gemma":[0.1187423,0.0002273132,0.0003927134,0.002337959,0.0003060066,0.0002333545,0.00007289927,0.001916951,0.000006408547],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006989909,"about_ca_system_score_gemma":0.007054273,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001043125,"about_ca_topic_score_gemma":0.00000472012,"domain_scores_codex":[0.991441,0.001627039,0.003991974,0.0003012081,0.002420265,0.0002185508],"domain_scores_gemma":[0.9915184,0.00183099,0.004824189,0.0004617463,0.001122552,0.0002421453],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000002615094,0.000135606,0.001237272,0.5532456,0.00008415229,0.000004109314,0.0007564266,1.698088e-8,6.282813e-8,0.0004621254,0.00131022,0.4427618],"study_design_scores_gemma":[0.0003459338,0.0001391237,0.05820716,0.9276198,0.003262632,0.0004124891,0.0004588611,0.00001091718,5.91773e-7,0.0003356501,0.009032581,0.0001742718],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.003190718,0.9858311,0.0001422534,0.006864436,0.00141479,0.002367543,0.000004208763,0.00001112116,0.000173822],"genre_scores_gemma":[0.01457228,0.9820839,0.0006394232,0.0008649383,0.001482791,0.0001108591,0.0001152367,0.0000476059,0.00008296312],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4425875,"threshold_uncertainty_score":0.9985748,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1412804313816833,"score_gpt":0.4201952112072738,"score_spread":0.2789147798255904,"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."}}