{"id":"W4360606116","doi":"10.1097/ju.0000000000003329.19","title":"MP66-19 INTER-OBSERVER RELIABILITY OF EXAMINER SCORING ON HIGH STAKES UROLOGY OBJECTIVE STRUCTURED CLINICAL EXAMINATION","year":2023,"lang":"en","type":"article","venue":"The Journal of Urology","topic":"Trauma and Emergency Care Studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Objective structured clinical examination; Medicine; Checklist; Summative assessment; Medical education; Context (archaeology); Internal consistency; Psychology; Formative assessment; Mathematics education; Psychometrics; Clinical psychology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01917088,0.0002561991,0.0004289044,0.001342667,0.0005100009,0.0007450869,0.0004900509,0.0004950947,0.01148917],"category_scores_gemma":[0.0562199,0.000316688,0.0007607482,0.0005786408,0.0004728457,0.0007583701,0.00179408,0.0005197917,0.003768583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004500753,"about_ca_system_score_gemma":0.0004546337,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00104343,"about_ca_topic_score_gemma":0.001984015,"domain_scores_codex":[0.9839535,0.00748133,0.001851124,0.001182407,0.005053928,0.0004776992],"domain_scores_gemma":[0.9433162,0.02746789,0.003999862,0.003714676,0.02017192,0.001329555],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003762253,0.0006951667,0.7827372,0.0006227258,0.000904535,0.0003794193,0.003646088,0.001337291,0.008069895,0.001050433,0.04908608,0.147709],"study_design_scores_gemma":[0.0001551623,0.001032682,0.9705282,0.0002845053,0.0001606554,0.0005705796,0.001045275,0.004789078,0.004415003,0.0004109495,0.01652887,0.00007898588],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9156423,0.001108028,0.02614265,0.0005739567,0.001169858,0.002149642,0.003520642,0.0007243397,0.04896857],"genre_scores_gemma":[0.9770108,0.000282921,0.01039921,0.0001389672,0.0001503342,0.001026908,0.001939656,0.0001996829,0.008851562],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01917088,"threshold_uncertainty_score":0.1013865,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06954669585200464,"score_gpt":0.3515792260740854,"score_spread":0.2820325302220807,"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."}}