{"id":"W2039358902","doi":"10.1080/01421590802334309","title":"Clerkship evaluation–what are we measuring?","year":2009,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Objective structured clinical examination; Principal (computer security); Educational measurement; Medical education; Multiple choice; Computer science; Psychology; Medicine; Curriculum; Pedagogy; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.1596602,0.001564474,0.002513126,0.005908472,0.002960762,0.01108224,0.003204684,0.002880598,0.00213169],"category_scores_gemma":[0.3455011,0.0008224781,0.001284595,0.008137926,0.008530289,0.01456494,0.003868592,0.004580562,0.0009551215],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01218856,"about_ca_system_score_gemma":0.01650393,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01452243,"about_ca_topic_score_gemma":0.01374678,"domain_scores_codex":[0.8509862,0.09984621,0.01362268,0.004799248,0.02700349,0.003742135],"domain_scores_gemma":[0.6393387,0.1722993,0.05700804,0.02073496,0.09415235,0.0164666],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001582611,0.0005951865,0.4697162,0.002652406,0.0003672283,0.00006461129,0.006558235,0.000865472,0.0003329228,0.007523422,0.02210551,0.4890606],"study_design_scores_gemma":[0.0001022873,0.001340421,0.8474939,0.02274189,0.0005623309,0.0006496154,0.03213206,0.005985482,0.001727019,0.03034485,0.0565768,0.0003433601],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.3977813,0.18098,0.07149416,0.2785266,0.005071348,0.003750382,0.002134991,0.001055239,0.05920601],"genre_scores_gemma":[0.93007,0.02281683,0.03418038,0.007204227,0.001258763,0.002584103,0.0007621246,0.0001626661,0.0009609314],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8403398,"threshold_uncertainty_score":0.8443742,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1147198611655236,"score_gpt":0.4022657078693233,"score_spread":0.2875458467037997,"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."}}