{"id":"W3164565344","doi":"10.1503/cjs.014919","title":"Assessing the acceptability of script concordance testing: a nationwide study in otolaryngology","year":2021,"lang":"en","type":"article","venue":"Canadian Journal of Surgery","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Medicine; Concordance; Test (biology); Otorhinolaryngology; Medical education; Family medicine; Medical physics; Surgery; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"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":[0.0161128,0.0002504999,0.0004733016,0.001747526,0.002338716,0.001846403,0.001162668,0.0009754958,0.001687905],"category_scores_gemma":[0.03865264,0.0007120309,0.0006136415,0.002155697,0.002534135,0.001816055,0.002405676,0.00169185,0.0003331272],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006191913,"about_ca_system_score_gemma":0.009660386,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1264256,"about_ca_topic_score_gemma":0.1789006,"domain_scores_codex":[0.9862835,0.004337016,0.001280966,0.001314902,0.005314505,0.001469086],"domain_scores_gemma":[0.9655446,0.008877051,0.00973534,0.001676494,0.0105868,0.003579666],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007887019,0.000329232,0.9667398,0.00008246172,0.0000308563,0.0002155812,0.02506824,0.00002875397,0.0005198373,0.00006409411,0.0004215825,0.006420778],"study_design_scores_gemma":[0.00001019744,0.0003547095,0.9666113,0.00007133382,0.00001390436,0.0002747842,0.03119726,0.0002583307,0.0001384774,0.00003322498,0.001012772,0.00002374149],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9990956,0.00005762541,0.0001197524,0.0001853051,0.00000568437,0.00006795814,0.00005290865,0.000002024964,0.0004131195],"genre_scores_gemma":[0.9988558,0.0001429479,0.0003143706,0.0003192962,0.000008162967,0.0001029236,0.00008981444,0.000006104598,0.0001605247],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1264256,"threshold_uncertainty_score":0.2513795,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1168689726064121,"score_gpt":0.3753410711170639,"score_spread":0.2584720985106518,"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."}}