{"id":"W4214632569","doi":"10.1016/j.jsurg.2022.01.014","title":"Validation of a Surgical Objective Structured Clinical Examination (S-OSCE) Using Convergent, Divergent, And Trainee-Based Assessments of Fidelity","year":2022,"lang":"en","type":"article","venue":"Journal of surgical education","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Objective structured clinical examination; Fidelity; Medical physics; Psychology; Medical education; Medicine; Computer science","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002092484,0.0001235209,0.0005387783,0.0002683359,0.00008899593,0.000009562892,0.00007085251,0.00009244798,0.001013209],"category_scores_gemma":[0.0002479847,0.000107701,0.0002560566,0.0004057168,0.0001099548,0.0001472177,0.00003540807,0.0003674628,2.104393e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001748376,"about_ca_system_score_gemma":0.0007531985,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002246329,"about_ca_topic_score_gemma":3.378594e-7,"domain_scores_codex":[0.9970074,0.0006574009,0.001282987,0.0001828218,0.0007448147,0.0001245364],"domain_scores_gemma":[0.9971365,0.0006494367,0.001202111,0.0001156785,0.0007016096,0.0001946837],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007217521,0.006801774,0.7942894,0.0003917415,0.0006126113,0.0001150934,0.001571691,0.00832012,0.005010747,0.001268425,0.0001187946,0.1742821],"study_design_scores_gemma":[0.01765845,0.002066637,0.9523698,0.0002051647,0.0005510239,0.0004406535,0.002646698,0.015553,0.002827906,0.0004100993,0.005035325,0.0002352207],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9978577,0.0001611348,0.0004389024,0.0001942106,0.0007343247,0.0002395814,0.0000250023,0.000006330321,0.00034282],"genre_scores_gemma":[0.9986867,0.0000579907,0.0009084411,0.0000444619,0.0001886932,0.000003106537,0.00006617262,0.00001101264,0.00003343793],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1740468,"threshold_uncertainty_score":0.9999,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08148684249735791,"score_gpt":0.4231819877449417,"score_spread":0.3416951452475838,"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."}}