{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01195411,0.0015187,0.007597373,0.01470117,0.0008623649,0.00308686,0.00246913,0.002090578,0.004646252],"category_scores_gemma":[0.09122422,0.001160081,0.006542348,0.01850331,0.00110482,0.003015355,0.001658311,0.00126616,0.0005179736],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00436475,"about_ca_system_score_gemma":0.01798347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01144062,"about_ca_topic_score_gemma":0.03124569,"domain_scores_codex":[0.9846911,0.003988544,0.006933126,0.001214491,0.002802129,0.0003707713],"domain_scores_gemma":[0.9080495,0.06852854,0.01460754,0.001134514,0.007086985,0.0005929791],"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":[0.0001204739,0.00001310478,0.002194299,0.9653122,0.006912943,0.0001691476,0.0002856494,0.00006487402,0.0001150029,0.0001726687,0.001665929,0.02297379],"study_design_scores_gemma":[0.0001118466,0.0001242953,0.009528629,0.9206682,0.05445888,0.0004876647,0.0003601101,0.00005356062,0.0001380911,0.000203384,0.01383426,0.00003111025],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001885847,0.99582,0.000186922,0.0003213801,0.0001559544,0.000286811,0.000936782,0.000009230072,0.0003969678],"genre_scores_gemma":[0.01818156,0.9789374,0.0007061717,0.0006320786,0.0001163866,0.0006906173,0.0005675391,0.00000848917,0.0001597041],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.01470117,"threshold_uncertainty_score":0.06322014,"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."}}