{"id":"W4249821236","doi":"10.1016/j.gie.2019.04.116","title":"804 DEVELOPMENT AND INITIAL VALIDATION OF AN INSTRUMENT FOR VIDEO-BASED ASSESSMENT OF TECHNICAL SKILL IN ERCP","year":2019,"lang":"en","type":"article","venue":"Gastrointestinal Endoscopy","topic":"Quality and Management Systems","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Credentialing; Coaching; Medicine; Quality assurance; Medical physics; Medical education; Rubric; Quality (philosophy); Quality assessment; Pathology; Psychology; Mathematics education","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":[0.01840518,0.0007658679,0.0005889359,0.001777079,0.0005924479,0.001180296,0.001307209,0.001290816,0.001655857],"category_scores_gemma":[0.02134508,0.0004693792,0.000895733,0.0007526647,0.000681842,0.0008661514,0.001521995,0.0008953379,0.000960395],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009909704,"about_ca_system_score_gemma":0.002201481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002451438,"about_ca_topic_score_gemma":0.002603762,"domain_scores_codex":[0.9910058,0.004386064,0.001046531,0.0005876654,0.002562043,0.0004118538],"domain_scores_gemma":[0.9819736,0.007561894,0.0008292267,0.0009996052,0.008155446,0.00048011],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002601512,0.008633171,0.5067246,0.0007670918,0.0002048394,0.0004755914,0.005446173,0.003104804,0.08813826,0.001461545,0.003245679,0.3791966],"study_design_scores_gemma":[0.0006531649,0.02421981,0.8409819,0.000506348,0.000303479,0.001107504,0.003986418,0.03580979,0.07770689,0.0007130302,0.0138438,0.0001678423],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.9130092,0.000247481,0.06959826,0.0003387391,0.000158057,0.01081524,0.001097291,0.0003450511,0.004390839],"genre_scores_gemma":[0.7551609,0.0003883938,0.2298297,0.0004368029,0.00008400607,0.008366264,0.002307626,0.0001140882,0.003312281],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.01840518,"threshold_uncertainty_score":0.09733707,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02833018047501314,"score_gpt":0.3006517016261248,"score_spread":0.2723215211511117,"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."}}