{"id":"W2005030293","doi":"10.5688/aj7410194","title":"Evaluation of Clinical Reasoning in Basic Emergencies Using a Script Concordance Test","year":2010,"lang":"en","type":"article","venue":"American Journal of Pharmaceutical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Concordance; Test (biology); Pharmacy; Medicine; Medical education; Multiple choice; Psychology; Family medicine; Significant difference; 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":[0.007031269,0.0001118061,0.0005781655,0.0001781409,0.00002540236,0.000009032478,0.000112407,0.00006328684,0.0002978391],"category_scores_gemma":[0.1641155,0.00009523953,0.0001809941,0.0005244163,0.0004654929,0.0001106491,0.00001803875,0.0008415317,0.000004644449],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009081729,"about_ca_system_score_gemma":0.00362166,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001501623,"about_ca_topic_score_gemma":0.0000144566,"domain_scores_codex":[0.9970955,0.0003428625,0.001362456,0.000173898,0.0008333443,0.0001919],"domain_scores_gemma":[0.9902648,0.005747599,0.001208052,0.0002266209,0.002094135,0.000458839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001835657,0.001560614,0.5053796,0.00001093688,0.00004086773,0.000004445421,0.0001945049,0.00003869104,0.01021362,0.0000742452,0.0002284184,0.4820705],"study_design_scores_gemma":[0.003906862,0.001544011,0.9491765,0.002138523,0.001472867,0.0004579454,0.002263645,0.0330767,0.004254422,0.0003994065,0.001111468,0.0001975982],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9956133,0.0004463982,0.0001467295,0.001118285,0.001888906,0.0001808972,0.00000183017,0.000005168025,0.0005984546],"genre_scores_gemma":[0.9891083,0.0001880362,0.009357207,0.0005785089,0.0007347361,0.000004235124,0.000002292958,0.00001363907,0.00001305097],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4818729,"threshold_uncertainty_score":0.8429255,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1336649764016181,"score_gpt":0.5336216288282879,"score_spread":0.3999566524266698,"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."}}