{"id":"W7110845057","doi":"10.36834/cmej.81428","title":"Assessment burden by design: exploring the variability in competence by design assessment forms","year":2025,"lang":"fr","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; University of Ottawa; York University; Toronto Metropolitan University; University of Toronto; University of Calgary; University of Saskatchewan","funders":"","keywords":"Milestone; Competence (human resources); Narrative; Context (archaeology); Presentation (obstetrics); Narrative review","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2658329,0.0006723169,0.0008263643,0.004646303,0.002277556,0.006239893,0.002068075,0.001162315,0.002047031],"category_scores_gemma":[0.5636193,0.0007649821,0.00157133,0.003884934,0.004464387,0.004578767,0.006008697,0.001960642,0.0003330152],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01031668,"about_ca_system_score_gemma":0.009339941,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005486833,"about_ca_topic_score_gemma":0.00733086,"domain_scores_codex":[0.6511841,0.2173573,0.03917477,0.01271749,0.07610351,0.003462815],"domain_scores_gemma":[0.2186929,0.6126831,0.08368699,0.03068657,0.05154904,0.002701396],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001530631,0.0005652892,0.6295163,0.001800878,0.000530996,0.0003101047,0.1333579,0.003217406,0.002092731,0.008551954,0.004928044,0.2135978],"study_design_scores_gemma":[0.0003453252,0.002999386,0.8258834,0.004395735,0.000378666,0.001157209,0.06739082,0.02753321,0.00467919,0.01829411,0.04647181,0.0004711459],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9149801,0.001449549,0.06853969,0.002154696,0.0002068519,0.002492896,0.000491326,0.0001989968,0.009485952],"genre_scores_gemma":[0.9723415,0.0001791289,0.02443422,0.0003589957,0.00004148233,0.001903889,0.0001973638,0.00009773878,0.0004457124],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9896833,"threshold_uncertainty_score":0.9053588,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03118776589560707,"score_gpt":0.3628048419039477,"score_spread":0.3316170760083406,"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."}}