{"id":"W2129940854","doi":"10.1016/j.jamcollsurg.2003.08.020","title":"Relationship between objective assessment of technical skills and subjective in-training evaluations in surgical residents","year":2003,"lang":"en","type":"article","venue":"Journal of the American College of Surgeons","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":155,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Medicine; Concurrent validity; Medical physics; Physical therapy; Psychometrics; Clinical psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00230452,0.0001021439,0.0007005341,0.000463578,0.00004947113,0.00000332841,0.00008152014,0.00005351082,0.00002441637],"category_scores_gemma":[0.003101382,0.00007474351,0.0001743709,0.001403644,0.0003334948,0.00007962934,0.00002901916,0.0004887744,1.48786e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002004468,"about_ca_system_score_gemma":0.000634137,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007157701,"about_ca_topic_score_gemma":0.00009398318,"domain_scores_codex":[0.9976869,0.0006063884,0.0008880284,0.0001217069,0.0005318634,0.0001650678],"domain_scores_gemma":[0.9942191,0.004465272,0.0008215792,0.0001427749,0.0002474064,0.0001038476],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001032697,0.0002177576,0.9969512,0.000008483853,0.00004305121,0.00004774515,0.0002679341,0.0003181695,0.00005813214,0.001579941,0.000005374724,0.0003989523],"study_design_scores_gemma":[0.002832522,0.0002512594,0.9931751,0.0002940879,0.00008416556,0.0001409886,0.001657782,0.00004848516,0.00008513691,0.001313076,0.00005361895,0.00006380636],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9942873,0.00004243018,0.00004189274,0.0004367608,0.00004821526,0.0002360662,0.00001798637,0.000002983668,0.00488641],"genre_scores_gemma":[0.9989004,0.00001627156,0.0009886138,0.00002246029,0.00001630497,0.000003191741,7.318175e-7,0.000009672476,0.00004237597],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004844034,"threshold_uncertainty_score":0.3712868,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04556634841595349,"score_gpt":0.3828702204317748,"score_spread":0.3373038720158213,"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."}}