{"id":"W2176361635","doi":"10.3402/meo.v20.29242","title":"Probing the effect of OSCE checklist length on inter-observer reliability and observer accuracy","year":2015,"lang":"en","type":"article","venue":"Medical Education Online","topic":"Innovations in Medical Education","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Checklist; Intraclass correlation; Objective structured clinical examination; Reliability (semiconductor); Medicine; Grading (engineering); Consistency (knowledge bases); Psychology; Medical education; Clinical psychology; Psychometrics; Computer science","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002790978,0.000180799,0.0003361505,0.000105738,0.00007366513,0.00001553725,0.0002338927,0.000189555,0.0003379513],"category_scores_gemma":[0.05493915,0.0001085731,0.00006085034,0.0005465113,0.0004299513,0.0001132059,0.00008108473,0.0007156608,0.00002196303],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002206767,"about_ca_system_score_gemma":0.001609088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001420126,"about_ca_topic_score_gemma":0.00001363284,"domain_scores_codex":[0.9976938,0.0002484583,0.0006244251,0.0003282737,0.000902837,0.0002021475],"domain_scores_gemma":[0.9976895,0.0005465882,0.0002308125,0.0006690745,0.0005618006,0.0003022095],"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":[0.0008052766,0.006705362,0.1062548,0.001224905,0.00009670646,0.00000380043,0.003705707,0.00001020062,0.0001619425,0.001630976,0.1439914,0.7354089],"study_design_scores_gemma":[0.01045994,0.005848888,0.5560041,0.004385788,0.0005485586,0.0002218386,0.005114981,0.01301027,0.004057807,0.001107494,0.3985024,0.0007379127],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9385664,0.0001723763,0.00009315024,0.0583512,0.001390155,0.0006837102,0.00000389833,0.00004384507,0.0006952803],"genre_scores_gemma":[0.9907445,0.00003242174,0.001377245,0.005633039,0.0008299737,0.00009630766,0.0002610721,0.0000229076,0.001002591],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.734671,"threshold_uncertainty_score":0.9530215,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02761102858187187,"score_gpt":0.374156929602999,"score_spread":0.3465459010211271,"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."}}