{"id":"W4391297309","doi":"10.2196/54401","title":"Development of a Clinical Simulation Video to Evaluate Multiple Domains of Clinical Competence: Cross-Sectional Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Cross-sectional study; Competence (human resources); Psychology; Computer science; Medicine; Social psychology; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01025021,0.0002018826,0.0007074486,0.0004594159,0.0001033269,0.00002874004,0.0002619794,0.0004332418,0.001244886],"category_scores_gemma":[0.02323512,0.0001710591,0.0002193859,0.001285014,0.0003526555,0.0001468474,0.00009898061,0.0008471318,0.0001497943],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001993883,"about_ca_system_score_gemma":0.01091986,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002946919,"about_ca_topic_score_gemma":0.00001782874,"domain_scores_codex":[0.9919304,0.0003261357,0.004324019,0.0006670198,0.002497395,0.0002550628],"domain_scores_gemma":[0.995795,0.001253311,0.0004906066,0.0005382726,0.00152551,0.0003973583],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003741441,0.006582958,0.7016663,0.0003686209,0.0001554612,0.000002764603,0.002749702,0.00005922271,0.00006659985,0.0002630067,0.002794572,0.2849167],"study_design_scores_gemma":[0.001851852,0.0005719848,0.9484824,0.0006174946,0.00007066101,0.00001001598,0.00102258,0.02511645,0.00004925385,0.00008569359,0.02197879,0.0001428376],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9858912,0.0000652966,0.004205344,0.001334079,0.006153159,0.001846712,0.000004082251,0.00008538215,0.0004147374],"genre_scores_gemma":[0.9727474,0.000007238316,0.02347753,0.001134341,0.001734268,0.0003751484,0.000166633,0.00003067257,0.0003267075],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2847738,"threshold_uncertainty_score":0.9996681,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1143727699796872,"score_gpt":0.5729542154600152,"score_spread":0.4585814454803281,"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."}}