{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01514931,0.0004290792,0.0005574682,0.001915595,0.0005423563,0.0008999171,0.0006459595,0.0005455168,0.001608866],"category_scores_gemma":[0.05704442,0.0003376029,0.0005739511,0.001271376,0.001036936,0.0007947156,0.001458414,0.0005103676,0.0004564394],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008926645,"about_ca_system_score_gemma":0.001289572,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003168294,"about_ca_topic_score_gemma":0.004095973,"domain_scores_codex":[0.9876087,0.004021701,0.001534285,0.001283319,0.005055936,0.0004961399],"domain_scores_gemma":[0.9526709,0.02163332,0.009009565,0.002375851,0.01187146,0.002439052],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005494174,0.0002060239,0.9680535,0.00006561266,0.0001206972,0.0001139265,0.002831555,0.0003272544,0.001250692,0.00009662547,0.0003958551,0.02598891],"study_design_scores_gemma":[0.00008017146,0.001194707,0.9912639,0.00005748188,0.00004772087,0.0008156686,0.001334158,0.002358937,0.001895468,0.0001157588,0.0008081393,0.00002799776],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9977233,0.00008222652,0.0008885568,0.00002878522,0.00002365266,0.00009889151,0.00007384556,0.00001411802,0.001066562],"genre_scores_gemma":[0.9972103,0.00005621957,0.002155295,0.00002027287,0.00001296002,0.00009971312,0.0002164711,0.000005713712,0.0002231133],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01514931,"threshold_uncertainty_score":0.08011818,"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."}}