{"id":"W2128894270","doi":"10.1186/1472-6947-5-18","title":"Online clinical reasoning assessment with the Script Concordance test: a feasibility study","year":2005,"lang":"en","type":"article","venue":"BMC Medical Informatics and Decision Making","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Test (biology); Concordance; Context (archaeology); Medicine; Health informatics; Web application; Summative assessment; Certification; Medical physics; Computer science; Medical education; World Wide Web; Formative assessment; Psychology; Pathology; Public health; Internal medicine","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"],"consensus_categories":[],"category_scores_codex":[0.004963698,0.0002167201,0.0006343738,0.0000558715,0.0002165918,0.0001129786,0.0002347748,0.0001622316,0.0001577688],"category_scores_gemma":[0.06336119,0.0001095375,0.0001179322,0.0002513322,0.0003356085,0.0001225826,0.0002572639,0.0008439587,0.00002022815],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005894414,"about_ca_system_score_gemma":0.0007022563,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000810013,"about_ca_topic_score_gemma":0.000199053,"domain_scores_codex":[0.9962333,0.0001144591,0.001476627,0.000276172,0.001576635,0.0003228471],"domain_scores_gemma":[0.9503288,0.04779668,0.0003886437,0.0006731781,0.0002349976,0.000577708],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002953535,0.001762041,0.5794523,0.00002549121,0.00003094593,0.00003187968,0.0002564284,0.00004526394,7.236034e-8,0.0001134978,0.002185173,0.4158015],"study_design_scores_gemma":[0.006517623,0.001906062,0.6284088,0.003894091,0.0001586212,0.000208929,0.003364705,0.3538338,2.625455e-7,0.0001368401,0.001390846,0.0001795041],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8726479,0.0001433019,0.1247999,0.0004802734,0.0001292807,0.0006506217,0.000006303932,0.00005712154,0.001085322],"genre_scores_gemma":[0.7535797,0.0001019701,0.2414973,0.004439327,0.0003142385,0.00001442823,0.00000661836,0.0000120466,0.00003442434],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.415622,"threshold_uncertainty_score":0.9445285,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08053502247139271,"score_gpt":0.4625996337942774,"score_spread":0.3820646113228847,"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."}}