{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001112348,0.0001547459,0.0003078541,0.0003021513,0.00003438076,0.00004597525,0.0001586814,0.00002707148,0.0000414619],"category_scores_gemma":[0.0001431409,0.0001536004,0.00004392479,0.0001936193,0.00005047189,0.0004622727,0.00009272568,0.00007825866,0.000005209917],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005956288,"about_ca_system_score_gemma":0.00007311133,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002374857,"about_ca_topic_score_gemma":0.00007454211,"domain_scores_codex":[0.9985785,0.00003166827,0.0006175894,0.000271855,0.000280504,0.0002198087],"domain_scores_gemma":[0.9991437,0.0001261163,0.0004117957,0.0001875563,0.0001150576,0.00001577702],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001331023,0.001782819,0.804978,0.007302947,0.00004521613,0.00001439462,0.000122235,0.003199313,0.1310739,0.04590367,0.0002406543,0.00400589],"study_design_scores_gemma":[0.02735,0.0026062,0.8131549,0.007946537,0.0002569701,0.00008826149,0.003098174,0.07734086,0.05572348,0.004204077,0.006552167,0.001678387],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9668154,0.000002577991,0.02945316,0.0001164963,0.0001038718,0.0009090769,0.00000510815,0.00003637156,0.002557925],"genre_scores_gemma":[0.8894458,1.262237e-7,0.1101969,0.0001325521,0.00006034159,0.00006690356,0.00007045722,0.00001394394,0.00001300955],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08074371,"threshold_uncertainty_score":0.6263642,"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."}}