{"id":"W2158828425","doi":"10.1111/j.1365-2923.2010.03863.x","title":"Script concordance testing: a review of published validity evidence","year":2011,"lang":"en","type":"review","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":170,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Centre de Santé et de Services Sociaux Cavendish; McGill University","funders":"","keywords":"Concordance; Psychology; External validity; Construct validity; Competence (human resources); PsycINFO; Content validity; Validity; Clinical psychology; Internal validity; MEDLINE; Social psychology; Applied psychology; Psychometrics; Medicine; Statistics; Mathematics","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003246014,0.0003256699,0.002573998,0.0001196182,0.00003036173,0.00001407472,0.0004364014,0.0004704903,0.0032748],"category_scores_gemma":[0.8236926,0.0002277971,0.0005078718,0.0008651994,0.0002765864,0.00009118969,0.00009642808,0.0008902242,0.000177449],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001463508,"about_ca_system_score_gemma":0.01876049,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002236591,"about_ca_topic_score_gemma":0.000001898567,"domain_scores_codex":[0.9961343,0.0003951872,0.001688178,0.0005869911,0.0009252177,0.0002701623],"domain_scores_gemma":[0.9436256,0.04663171,0.002935354,0.00268683,0.002447149,0.001673387],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003275338,0.000454291,0.0001430519,0.1187024,0.00002472098,0.00000478391,0.000008832647,2.024538e-10,9.291671e-9,0.00006941958,0.1584959,0.7220933],"study_design_scores_gemma":[0.00007018021,0.0001236999,0.00009973403,0.499073,0.001009526,0.00010873,0.000001505575,7.211507e-7,1.412121e-7,0.00004130763,0.4993751,0.00009640254],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000004035141,0.9902592,0.00003548689,0.001978917,0.001832844,0.001106357,0.000006240021,0.00005844329,0.004718492],"genre_scores_gemma":[0.000005717808,0.9903753,0.001186392,0.006196987,0.0008012505,0.0002914553,0.0002110709,0.00003638123,0.0008954693],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8204466,"threshold_uncertainty_score":0.9976363,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2272363726862678,"score_gpt":0.4794160776581745,"score_spread":0.2521797049719066,"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."}}