{"id":"W3161895874","doi":"10.1080/0142159x.2021.1925642","title":"On the validity of summative entrustment decisions","year":2021,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":34,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of Toronto; University of Ottawa; Medical Council of Canada","funders":"","keywords":"Summative assessment; Argument (complex analysis); Psychology; Process (computing); Health care; External validity; Nursing; Medical education; Formative assessment; Medicine; Social psychology; Computer science; Pedagogy; Political science","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":[0.5323914,0.001235179,0.002253011,0.008255748,0.007283696,0.01184101,0.006829742,0.009198425,0.003612299],"category_scores_gemma":[0.867355,0.0008668148,0.002875577,0.004247309,0.05392445,0.01676637,0.01295359,0.01036466,0.0009094298],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01034409,"about_ca_system_score_gemma":0.01050936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005521561,"about_ca_topic_score_gemma":0.003285507,"domain_scores_codex":[0.3906123,0.4259368,0.03490321,0.02070925,0.1180233,0.009815123],"domain_scores_gemma":[0.06346814,0.8414395,0.0304297,0.02720175,0.03525722,0.002203625],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.004459268,0.0006275602,0.107752,0.001635869,0.001405521,0.001413557,0.07276212,0.007283013,0.0009342271,0.6357918,0.0154002,0.150535],"study_design_scores_gemma":[0.0006378451,0.0009715941,0.04590475,0.004879448,0.0005742303,0.00066297,0.02145502,0.03072471,0.006716526,0.8586711,0.02815306,0.0006487225],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4533461,0.006416237,0.1582992,0.1954597,0.004812049,0.00304148,0.0006836806,0.0004079425,0.1775336],"genre_scores_gemma":[0.9635783,0.0003801154,0.02372468,0.008384608,0.0007452295,0.001211843,0.0001496155,0.00006617796,0.001759406],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5323914,"threshold_uncertainty_score":0.5766447,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09308446346101831,"score_gpt":0.3871435513230113,"score_spread":0.294059087861993,"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."}}