{"id":"W2130963076","doi":"10.1016/j.jamcollsurg.2014.03.051","title":"A Global Delphi Consensus Study on Defining and Measuring Quality in Surgical Training","year":2014,"lang":"en","type":"article","venue":"Journal of the American College of Surgeons","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":42,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"National Institute for Health and Care Research","keywords":"Medicine; Trainer; Delphi method; Quality (philosophy); Likert scale; Delphi; Cronbach's alpha; Medical education; Scale (ratio); Medical physics; Psychology; Psychometrics; Computer science; Artificial intelligence","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":[0.00264192,0.000127175,0.0008109175,0.0001406214,0.00007213167,0.000006998305,0.00009747835,0.00002844972,0.00001033212],"category_scores_gemma":[0.00156706,0.00008462633,0.0001949823,0.0006557354,0.0002406025,0.00002388545,0.0000419116,0.0002704924,4.738949e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007611734,"about_ca_system_score_gemma":0.0001584749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001435362,"about_ca_topic_score_gemma":0.0001140715,"domain_scores_codex":[0.9977428,0.000570658,0.0007946977,0.0001381675,0.0005450849,0.0002086021],"domain_scores_gemma":[0.9969696,0.001822116,0.0007260679,0.0001690291,0.0001414288,0.0001717472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002013695,0.0004002863,0.9794061,0.00001114904,0.0001241461,0.0002574924,0.0006241612,0.0006253469,0.00002433599,0.001110361,0.00001632475,0.01538659],"study_design_scores_gemma":[0.005880188,0.0007921057,0.9814256,0.0002451543,0.00006768847,0.0004798719,0.009686028,0.000415286,0.00001014792,0.0001494869,0.0007538507,0.00009462873],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.994978,0.00004217588,0.00001105312,0.001136187,0.0001055083,0.0001428474,0.00001230773,0.000007640042,0.003564332],"genre_scores_gemma":[0.9995825,0.00000506056,0.0001872254,0.0001673295,0.00003587567,0.000001066292,1.461558e-7,0.000008844387,0.00001190898],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01529196,"threshold_uncertainty_score":0.3450961,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07474428196894926,"score_gpt":0.3482183023058162,"score_spread":0.2734740203368669,"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."}}