{"id":"W2996540752","doi":"10.1007/s40670-019-00855-6","title":"Daily Evaluation Cards Are Superior for Student Assessment Compared to Single Rater In-Training Evaluations","year":2019,"lang":"en","type":"article","venue":"Medical Science Educator","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Manitoba; Health Sciences Centre","funders":"","keywords":"Cohort; Medicine; Multivariate statistics; Ambulatory; Rating scale; Physical therapy; Medical physics; Medical education; Statistics; Internal medicine; Mathematics","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009106869,0.0001578649,0.0003539207,0.0005461233,0.0002080689,0.00009024737,0.0004304787,0.0001034642,0.002653801],"category_scores_gemma":[0.007896723,0.0001329007,0.00005345073,0.001974093,0.0003086803,0.0002220137,0.00008615006,0.0003043615,0.0001270641],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001978514,"about_ca_system_score_gemma":0.009132273,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003152878,"about_ca_topic_score_gemma":0.00003842911,"domain_scores_codex":[0.9942194,0.00009792637,0.0006013539,0.0006205278,0.003888932,0.0005718318],"domain_scores_gemma":[0.9975815,0.0001580067,0.0001106801,0.0004857799,0.00120684,0.0004572021],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001000913,0.004135496,0.689958,0.0002218023,0.00005677109,0.00000516385,0.03168696,0.0003222185,0.03603398,0.00140881,0.02516238,0.2109083],"study_design_scores_gemma":[0.002626562,0.0004322004,0.929759,0.0005563684,0.00007700775,0.00001598517,0.03222714,0.02254794,0.0007531981,0.0001398927,0.01060293,0.0002617189],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9631137,0.00002524647,0.0009844668,0.02709109,0.004520817,0.002725034,0.000003037292,0.00003159182,0.001505041],"genre_scores_gemma":[0.9853544,8.915451e-7,0.008171374,0.004576613,0.0005693073,0.0009817848,0.00005422943,0.00001682744,0.0002746361],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.239801,"threshold_uncertainty_score":0.9982579,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09964731443661813,"score_gpt":0.5047773264578296,"score_spread":0.4051300120212115,"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."}}