{"id":"W4391112649","doi":"10.5334/pme.1110","title":"The Next Era of Assessment: Building a Trustworthy Assessment System","year":2024,"lang":"en","type":"article","venue":"Perspectives on Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; University of Calgary","funders":"","keywords":"Contextualization; Context (archaeology); Variety (cybernetics); Medical education; Trustworthiness; Knowledge management; Computer science; Psychology; Public relations; Engineering ethics; Medicine; Political science; Social psychology; Interpretation (philosophy)","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":[0.1661121,0.0009717112,0.00170334,0.01074883,0.01414022,0.0264481,0.005216937,0.008442517,0.003949543],"category_scores_gemma":[0.2125417,0.001856009,0.002086712,0.003994021,0.0191342,0.04607989,0.03016644,0.01126158,0.002747002],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01141181,"about_ca_system_score_gemma":0.0366453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008203274,"about_ca_topic_score_gemma":0.005640392,"domain_scores_codex":[0.8407387,0.09817386,0.01702352,0.01312487,0.02684146,0.004097682],"domain_scores_gemma":[0.7232953,0.1154767,0.02418819,0.04083523,0.07567049,0.02053405],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002105455,0.0006585528,0.05330589,0.00101657,0.000390399,0.001481718,0.07045838,0.008369536,0.001859356,0.4056623,0.02929864,0.4272882],"study_design_scores_gemma":[0.0001335932,0.0007277038,0.01529468,0.005060179,0.00031602,0.0011832,0.03539768,0.05330518,0.002410887,0.6672459,0.218099,0.000826036],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"commentary","genre_scores_codex":[0.04761311,0.003198781,0.739494,0.1297394,0.001448149,0.002738929,0.0002179743,0.002431412,0.07311823],"genre_scores_gemma":[0.4362102,0.001269556,0.5509188,0.004628901,0.0004914928,0.00131341,0.0003153778,0.0002875708,0.004564692],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.1661121,"threshold_uncertainty_score":0.8784952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01414073531963129,"score_gpt":0.4025149743163983,"score_spread":0.388374238996767,"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."}}