{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02278472,0.0007712285,0.002066356,0.001903998,0.0005622327,0.002711137,0.001086997,0.001640969,0.01184202],"category_scores_gemma":[0.1481148,0.0005006844,0.0012829,0.001304111,0.0003609379,0.003441761,0.001402627,0.0009438485,0.004797439],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005767068,"about_ca_system_score_gemma":0.0008964741,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001185326,"about_ca_topic_score_gemma":0.001972336,"domain_scores_codex":[0.9744665,0.01372799,0.002917204,0.001663901,0.006779333,0.0004452292],"domain_scores_gemma":[0.7802508,0.1660209,0.0169867,0.01157429,0.01917268,0.005994663],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.03991361,0.006286626,0.2490694,0.001323908,0.001101945,0.0001230425,0.0008414822,0.001735772,0.003257653,0.0004643378,0.01961315,0.676269],"study_design_scores_gemma":[0.004011295,0.02989572,0.9196162,0.0008856259,0.001144882,0.0008118714,0.001175104,0.01565381,0.007639362,0.001488169,0.01727716,0.000400858],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9506896,0.003862834,0.01149903,0.0008721682,0.001810276,0.001272442,0.001832229,0.001681623,0.02647991],"genre_scores_gemma":[0.966486,0.001717113,0.01713081,0.0005807648,0.0003673136,0.0004979954,0.002106197,0.0002705922,0.01084326],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02278472,"threshold_uncertainty_score":0.1204986,"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."}}