{"id":"W2131996163","doi":"10.36834/cmej.36531","title":"The Reliability and Validity of a Paediatric Script Concordance Test with Medical Students, Paediatric Residents and Experienced Paediatricians","year":2010,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; University of Calgary","funders":"Royal College of Physicians and Surgeons of Canada","keywords":"Concordance; Test (biology); Grading (engineering); Reliability (semiconductor); Medicine; Multiple choice; Medical education; Family medicine; Psychology; Pediatrics; Significant difference; Internal medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003957607,0.0001751782,0.0003588384,0.000180504,0.0004067881,0.0001176192,0.0004134638,0.0003167548,0.001124295],"category_scores_gemma":[0.5463667,0.0001094973,0.00006457268,0.000470841,0.00100147,0.00008670689,0.00006207589,0.001528751,0.000004775037],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001000158,"about_ca_system_score_gemma":0.02983006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004834711,"about_ca_topic_score_gemma":0.0146203,"domain_scores_codex":[0.9962957,0.0001690889,0.0007054721,0.000322833,0.002067302,0.0004395958],"domain_scores_gemma":[0.9621365,0.01844848,0.0005036601,0.0006244964,0.001072634,0.01721424],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004885594,0.0002847868,0.9164519,0.00002267931,0.00001642854,0.00009406854,0.0004397578,1.014891e-7,0.000002608162,0.00008046789,0.02508399,0.05747434],"study_design_scores_gemma":[0.002176926,0.0005851912,0.9822631,0.0005521825,0.0001545838,0.001948523,0.001417491,0.00009512861,0.000009343182,0.0003386907,0.0102802,0.0001785873],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9621051,0.001755198,0.00001873152,0.03407845,0.001518031,0.0002690585,0.000008394942,0.00001172343,0.0002353231],"genre_scores_gemma":[0.9871163,0.004231481,0.0003560474,0.006500172,0.001598631,0.00002649282,0.00000576027,0.00001715755,0.0001480088],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5424091,"threshold_uncertainty_score":0.9997888,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009798068903059506,"score_gpt":0.3212783576356686,"score_spread":0.3114802887326091,"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."}}