{"id":"W3096953036","doi":"10.1080/0142159x.2020.1841128","title":"Exploring the connections between programmatic assessment and program evaluation within competency-based medical education programs","year":2020,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Medical education; Program evaluation; Competency assessment; Evaluation methods; Self-assessment; Computer science; Medicine; Psychology; Political science; Pedagogy; Engineering","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3055108,0.0006941137,0.001297648,0.007939226,0.00537175,0.01378835,0.002872934,0.002315733,0.001787701],"category_scores_gemma":[0.47232,0.0009027949,0.0007694392,0.004621011,0.01110061,0.01304634,0.01424936,0.005659485,0.0002094549],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02128714,"about_ca_system_score_gemma":0.0598876,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01009838,"about_ca_topic_score_gemma":0.01610372,"domain_scores_codex":[0.525956,0.4274242,0.01111952,0.005009535,0.02578618,0.004704667],"domain_scores_gemma":[0.3774726,0.5226379,0.0329793,0.01052241,0.04643737,0.009950485],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003962447,0.001615642,0.07376159,0.002360615,0.0002996658,0.0001952272,0.03786029,0.005304704,0.0004282529,0.179222,0.003293061,0.6952627],"study_design_scores_gemma":[0.0009688258,0.005082773,0.2944895,0.02374214,0.0006809703,0.001087079,0.07762803,0.05390218,0.006764846,0.4466377,0.08806631,0.0009497142],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2541836,0.02086003,0.4425697,0.1415487,0.001429181,0.005840741,0.0001635078,0.0008740586,0.1325304],"genre_scores_gemma":[0.9140394,0.001610038,0.07879781,0.002187359,0.000184933,0.002071962,0.00004000101,0.00007227777,0.0009963916],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3055108,"threshold_uncertainty_score":0.8564289,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1869079731289644,"score_gpt":0.4421498276901076,"score_spread":0.2552418545611432,"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."}}