{"id":"W2060385292","doi":"10.1186/1472-6920-12-123","title":"“Who writes what?” Using written comments in team-based assessment to better understand medical student performance: a mixed-methods study","year":2012,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Excellence; Medical education; Quality (philosophy); Health care; Psychology; Medicine","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008560215,0.0003077154,0.000548654,0.0007247566,0.000186379,0.00008722136,0.000380916,0.000359004,0.001966243],"category_scores_gemma":[0.005123632,0.0002703124,0.00007119973,0.001340905,0.0001732142,0.0004838,0.0001562414,0.0009811405,0.00004207654],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001634802,"about_ca_system_score_gemma":0.006459997,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001010309,"about_ca_topic_score_gemma":0.00005078998,"domain_scores_codex":[0.9937029,0.0008712194,0.001143972,0.0004868172,0.003101951,0.0006930706],"domain_scores_gemma":[0.9975498,0.0004666948,0.0001908251,0.0006288003,0.0003076738,0.0008561821],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005083327,0.009934576,0.7700129,0.0002772971,0.00004325182,0.00000201422,0.003798693,0.000008137254,0.00002245016,0.00008739797,0.01078448,0.204978],"study_design_scores_gemma":[0.003651941,0.0005610779,0.8969836,0.002788144,0.0001651871,0.00004307385,0.06752902,0.02410263,0.00007383019,0.00003293631,0.003675222,0.0003933709],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9487137,0.0002162034,0.03112995,0.01374719,0.004045538,0.00169398,5.749015e-7,0.00005232252,0.0004005272],"genre_scores_gemma":[0.8995,0.00003609041,0.07179955,0.02677225,0.001250037,0.0003975612,0.0001043572,0.00004639533,0.00009383539],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2045846,"threshold_uncertainty_score":0.9999749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04911330270045402,"score_gpt":0.502360400308022,"score_spread":0.453247097607568,"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."}}