{"id":"W2052585986","doi":"10.1016/s0002-9610(02)00798-5","title":"A role for error training in surgical technical skill instruction and evaluation","year":2002,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":72,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Inter-rater reliability; Rating scale; Psychology; Dreyfus model of skill acquisition; Reliability (semiconductor); Computer science; Medical education; Applied psychology; Medicine; Developmental psychology","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":[0.02655281,0.001061945,0.0008717127,0.002472698,0.00271477,0.007223864,0.002663089,0.004809522,0.01579117],"category_scores_gemma":[0.1206834,0.0006341696,0.0008737676,0.0007795303,0.004501225,0.009384726,0.003949061,0.00361568,0.003055593],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001911802,"about_ca_system_score_gemma":0.01066781,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006625014,"about_ca_topic_score_gemma":0.006531915,"domain_scores_codex":[0.9809198,0.01053305,0.001118168,0.001369737,0.005330142,0.0007291991],"domain_scores_gemma":[0.7754408,0.1601028,0.007416443,0.01551591,0.03481229,0.00671175],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001031003,0.001161283,0.02897377,0.001106087,0.0001059635,0.0004644086,0.004540193,0.00265393,0.004330255,0.03658221,0.01570695,0.903344],"study_design_scores_gemma":[0.0008944609,0.005945555,0.1480172,0.01478172,0.0008366652,0.01616001,0.01750113,0.103761,0.04245368,0.2797732,0.3687301,0.001145464],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1529666,0.03070066,0.4667851,0.1223712,0.008958495,0.0006573061,0.0003461408,0.005520182,0.2116943],"genre_scores_gemma":[0.6811689,0.006009986,0.2710215,0.007639041,0.001976673,0.0004008441,0.0001026937,0.0008570468,0.03082328],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02655281,"threshold_uncertainty_score":0.1404263,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1042056931113103,"score_gpt":0.3498988343217001,"score_spread":0.2456931412103898,"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."}}