{"id":"W2064656139","doi":"10.1097/acm.0000000000000704","title":"Diagnosing Technical Competence in Six Bedside Procedures","year":2015,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":66,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Checklist; Inter-rater reliability; Competence (human resources); Formative assessment; Rating scale; Psychology; Kappa; Clinical psychology; Medicine; Social psychology; Developmental psychology; Mathematics","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":[0.0009810516,0.000164524,0.0004092266,0.0003139781,0.00003903759,0.000002379487,0.0001643091,0.0002846447,0.0001360793],"category_scores_gemma":[0.005915229,0.0001335988,0.00002172129,0.0006671801,0.0002342616,0.00009369932,0.00003343117,0.001005646,0.00006590373],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004169621,"about_ca_system_score_gemma":0.001027445,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003918852,"about_ca_topic_score_gemma":0.00003705818,"domain_scores_codex":[0.9979066,0.0000762268,0.0007179223,0.0003413399,0.0006090506,0.0003488564],"domain_scores_gemma":[0.9977977,0.0009339937,0.0001451305,0.0003124358,0.000297352,0.0005134124],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008171488,0.00002145758,0.9596444,0.0001688711,0.000004121773,0.00002605371,0.004861952,0.00009383423,0.0006527309,0.0008480775,0.0329668,0.0006299571],"study_design_scores_gemma":[0.006502917,0.0006922847,0.9515588,0.005875075,0.0001051601,0.0005360577,0.01486228,0.001694434,0.0009822141,0.005250868,0.01154429,0.0003955762],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9040638,0.004098972,0.0001626203,0.08773277,0.0006862064,0.0008235661,0.000001277433,0.0001761615,0.002254693],"genre_scores_gemma":[0.9877519,0.0001897746,0.0005310394,0.00988913,0.001243718,0.00009736182,0.00004298086,0.00002623023,0.0002278864],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08368815,"threshold_uncertainty_score":0.7081509,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1246061693202169,"score_gpt":0.4426169102860012,"score_spread":0.3180107409657843,"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."}}