{"id":"W2150819136","doi":"10.1080/01421590802512938","title":"Assessing postgraduate trainees in Canada: Are we achieving diversity in methods?","year":2008,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Logbook; Accreditation; Medical education; Specialty; Medicine; Diversity (politics); Program evaluation; Family medicine; Educational measurement; Curriculum; MEDLINE; Psychology","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001746141,0.000131572,0.0003579098,0.0002321941,0.0001182236,0.000006073998,0.0001697562,0.000158604,0.0009535807],"category_scores_gemma":[0.005745801,0.0001153976,0.00003529732,0.0007746856,0.0001437859,0.0001262712,0.0001232067,0.00114961,0.00000495867],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001287489,"about_ca_system_score_gemma":0.002067845,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2935425,"about_ca_topic_score_gemma":0.1954393,"domain_scores_codex":[0.997804,0.0002734234,0.0004312939,0.0002609523,0.00088969,0.0003406313],"domain_scores_gemma":[0.9993867,0.000090503,0.00008673945,0.0002033606,0.00006333687,0.0001693812],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001180389,0.0002144414,0.912526,0.00004865789,0.00001205193,0.0005082486,0.002347191,0.000002117105,0.00002040068,0.00002206597,0.005956376,0.07833067],"study_design_scores_gemma":[0.00122253,0.00001795024,0.9813551,0.0005410012,0.00001584182,0.000133696,0.005406078,0.007303532,0.00002013269,0.00007620006,0.00375553,0.0001523603],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9565169,0.000224673,0.001184363,0.03968553,0.0003373305,0.0001578046,2.917345e-7,0.00002679298,0.001866347],"genre_scores_gemma":[0.9914349,0.00007400494,0.005846178,0.002010425,0.00019171,0.00001126613,0.00001346051,0.00001462277,0.0004034795],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09810314,"threshold_uncertainty_score":0.9999597,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1344025216552011,"score_gpt":0.4179195934100754,"score_spread":0.2835170717548743,"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."}}