{"id":"W2315264017","doi":"10.1097/sla.0000000000001620","title":"Implementing and Evaluating a National Certification Technical Skills Examination","year":2016,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":61,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Sinai Hospital","funders":"","keywords":"Certification; Medicine; Cronbach's alpha; Competence (human resources); Checklist; Reliability (semiconductor); Rating scale; Maintenance of Certification; Medical education; Medical physics; Psychometrics; Clinical psychology; Psychology; Management","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":[0.0354187,0.000390443,0.0004915347,0.001048421,0.000829243,0.001006112,0.000782095,0.0004808446,0.001453875],"category_scores_gemma":[0.06196453,0.000231204,0.0004109469,0.0005151432,0.0004820344,0.0008254337,0.001724566,0.0005780355,0.0005817974],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003278079,"about_ca_system_score_gemma":0.01645554,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009079176,"about_ca_topic_score_gemma":0.02076579,"domain_scores_codex":[0.9805678,0.00922837,0.00174525,0.0009560963,0.006055932,0.001446577],"domain_scores_gemma":[0.9501191,0.008352471,0.005881357,0.00204594,0.02951078,0.004090284],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006378611,0.003840783,0.6874495,0.0003288319,0.0000724271,0.0001336232,0.001630913,0.002235244,0.002725308,0.0002726142,0.003559332,0.2971136],"study_design_scores_gemma":[0.0001071949,0.006413986,0.9767248,0.0002380955,0.00006352996,0.000142376,0.00240878,0.004772266,0.003390346,0.00008423924,0.005621665,0.00003267567],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9857637,0.0001928482,0.003344712,0.000512582,0.00008524671,0.002029798,0.0001667705,0.0001256391,0.007778686],"genre_scores_gemma":[0.9762211,0.0001990057,0.0192923,0.0001689231,0.00002439042,0.0009639707,0.000834559,0.00002005968,0.002275692],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0354187,"threshold_uncertainty_score":0.1873143,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4304403721875084,"score_gpt":0.4635618905539116,"score_spread":0.03312151836640315,"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."}}