{"id":"W3157388129","doi":"10.1016/j.jss.2021.03.057","title":"The Impact of Surgeon Experience on Script Concordance Test Scoring","year":2021,"lang":"en","type":"article","venue":"Journal of Surgical Research","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa","funders":"University of Ottawa","keywords":"Concordance; Medicine; Test (biology); Specialty; Subgroup analysis; Family medicine; Meta-analysis; Internal medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03344157,0.0004705826,0.0007030523,0.001536316,0.001164163,0.002855142,0.001316424,0.0008873449,0.004555167],"category_scores_gemma":[0.2869756,0.0004780644,0.001017073,0.001487855,0.001061738,0.002163717,0.001746592,0.00117119,0.0008495415],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001685459,"about_ca_system_score_gemma":0.002993589,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004505192,"about_ca_topic_score_gemma":0.01441933,"domain_scores_codex":[0.955938,0.02425284,0.005734984,0.002963885,0.009780212,0.001330075],"domain_scores_gemma":[0.412613,0.5145233,0.02467798,0.01190466,0.02771429,0.008566815],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006428277,0.001090867,0.8208609,0.000254167,0.0005407404,0.0007158599,0.005349419,0.002196968,0.002619138,0.0003355393,0.002908145,0.1567],"study_design_scores_gemma":[0.0001748128,0.00646969,0.9639922,0.0002749491,0.0006654578,0.002769158,0.004482394,0.01288396,0.00394795,0.0007679068,0.003381243,0.0001902364],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9909574,0.0006526065,0.00230662,0.0005683606,0.0002911594,0.00005898007,0.0001802501,0.00009409497,0.004890529],"genre_scores_gemma":[0.9963087,0.0001548111,0.002214137,0.0001704567,0.0000946326,0.00003360854,0.0001421305,0.00009398773,0.0007874845],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9665585,"threshold_uncertainty_score":0.176858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1482298942759711,"score_gpt":0.5095313188142345,"score_spread":0.3613014245382634,"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."}}