{"id":"W2155284013","doi":"10.3109/0142159x.2013.760036","title":"Script concordance testing: From theory to practice: AMEE Guide No. 75","year":2013,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":190,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Concordance; Competence (human resources); Medical education; Psychology; Test (biology); Variety (cybernetics); Medicine; Applied psychology; Management science; Computer science; Social psychology; Artificial intelligence","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":["insufficient_payload"],"category_scores_codex":[0.001641395,0.0002127932,0.0004807578,0.00003990886,0.00005929939,0.00004200178,0.0002606522,0.00027623,0.02192793],"category_scores_gemma":[0.6780049,0.0001545437,0.0001019321,0.0002157513,0.000237443,0.00009070096,0.0001432264,0.0008095776,0.01310986],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007746404,"about_ca_system_score_gemma":0.0003210803,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001723533,"about_ca_topic_score_gemma":0.000005924771,"domain_scores_codex":[0.9971939,0.0002651883,0.0005798086,0.000500883,0.001008289,0.0004519395],"domain_scores_gemma":[0.959431,0.03567024,0.0002213975,0.001136752,0.0008527678,0.00268783],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001403879,0.000771646,0.06795765,0.000009645696,0.0001246289,0.0002734477,0.0003066287,4.927101e-7,0.0002028206,0.0001952483,0.7252444,0.204773],"study_design_scores_gemma":[0.003418652,0.0009183047,0.06537399,0.003186094,0.0003287945,0.00009795817,0.0006643305,0.001900871,0.00007914047,0.002877891,0.9207127,0.0004412386],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.582793,0.001242142,0.009270092,0.05797454,0.002845478,0.001248448,0.000009090505,0.0006060692,0.3440111],"genre_scores_gemma":[0.5113795,0.000127017,0.1083331,0.1541424,0.008660712,0.0004605853,0.0001043427,0.0001888583,0.2166036],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6763635,"threshold_uncertainty_score":0.9876586,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03335734072977935,"score_gpt":0.3691587632072861,"score_spread":0.3358014224775068,"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."}}