{"id":"W2588982045","doi":"10.1097/acm.0000000000001559","title":"Transforming Medical Assessment: Integrating Uncertainty Into the Evaluation of Clinical Reasoning in Medical Education","year":2017,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":77,"is_retracted":false,"has_abstract":true,"ca_institutions":"Royal College of Physicians and Surgeons of Canada; University of Calgary","funders":"","keywords":"Context (archaeology); MEDLINE; Engineering ethics; Management science; Field (mathematics); Psychology; Medical education; Medicine; Knowledge management; Computer science; 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":[0.09727318,0.001060489,0.001082784,0.008308444,0.003832486,0.02434061,0.002991247,0.004932829,0.0022598],"category_scores_gemma":[0.204649,0.0008640523,0.001318375,0.003577879,0.03327449,0.02834746,0.01696749,0.01100159,0.0004942792],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01071329,"about_ca_system_score_gemma":0.02375813,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004929223,"about_ca_topic_score_gemma":0.007457879,"domain_scores_codex":[0.8914162,0.08774357,0.005428098,0.001760827,0.01194438,0.001706839],"domain_scores_gemma":[0.7518619,0.2041777,0.01140863,0.00790263,0.01831246,0.006336624],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001451388,0.0002208241,0.0115826,0.001085931,0.0001019883,0.0003267445,0.04680431,0.004177073,0.000489036,0.5487383,0.00867425,0.3776538],"study_design_scores_gemma":[0.00003109239,0.0001181911,0.00331598,0.002548441,0.0000549506,0.000466516,0.01765934,0.006684331,0.0007025458,0.9248748,0.0434105,0.0001334291],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03917011,0.01978974,0.6005474,0.2285518,0.002625052,0.0008995121,0.0001172477,0.0007443136,0.1075549],"genre_scores_gemma":[0.5740177,0.006804858,0.4041143,0.01120892,0.0008820749,0.0006738686,0.00006930501,0.0001214191,0.002107482],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.09727318,"threshold_uncertainty_score":0.5144359,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1109682342709774,"score_gpt":0.5498533590944679,"score_spread":0.4388851248234905,"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."}}