{"id":"W2917561174","doi":"10.1111/medu.13814","title":"Experts’ responses in script concordance tests: a response process validity investigation","year":2019,"lang":"en","type":"article","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"College of Family Physicians of Canada","funders":"","keywords":"Concordance; Context (archaeology); Consistency (knowledge bases); Test (biology); Psychology; Process (computing); Action (physics); Qualitative property; Applied psychology; Medicine; Social psychology; Computer science; Artificial intelligence; Machine learning","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":[0.001905226,0.0001218172,0.0002753282,0.0001466843,0.00002895422,0.00001706159,0.0001114764,0.0001924874,0.0007605194],"category_scores_gemma":[0.3363081,0.0001033425,0.00004642298,0.0004199618,0.0001508098,0.00009455466,0.00002021583,0.0002917759,0.0001775846],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000171969,"about_ca_system_score_gemma":0.00501436,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001623422,"about_ca_topic_score_gemma":0.00002452728,"domain_scores_codex":[0.9979987,0.0003453839,0.0004352317,0.0003493803,0.0006534702,0.0002178622],"domain_scores_gemma":[0.9896086,0.009254192,0.0001227856,0.0003930102,0.0002180543,0.0004033731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002525977,0.0009839227,0.9593017,0.00008079535,0.000006111387,0.00002255326,0.001930991,0.000001334095,0.001095228,0.0001133914,0.02143259,0.01250537],"study_design_scores_gemma":[0.00219029,0.0005951999,0.980946,0.005360806,0.00002085938,0.00007512316,0.001267066,0.0005068798,0.001328409,0.00098994,0.006542751,0.00017669],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9718125,0.0002515746,0.0000180427,0.02580314,0.0008713999,0.0004501397,0.00000152501,0.00006412478,0.0007275089],"genre_scores_gemma":[0.9830665,0.00004384769,0.0002974561,0.01381432,0.0002251274,0.0001304168,0.00004375787,0.00001577886,0.002362772],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3344029,"threshold_uncertainty_score":0.8895261,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04101367176045202,"score_gpt":0.402577520766441,"score_spread":0.361563849005989,"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."}}