{"id":"W2795046142","doi":"10.36834/cmej.42286","title":"Redeveloping a workplace-based assessment program for physicians using Kane’s validity framework","year":2018,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"College of Physicians and Surgeons of Ontario","funders":"","keywords":"Formative assessment; Process (computing); Specialty; Medical education; Computer science; Knowledge management; Psychology; Medicine; Family medicine; Pedagogy","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00167493,0.000187594,0.0002820065,0.0005346764,0.0007695476,0.0001283163,0.00025009,0.000388081,0.006793614],"category_scores_gemma":[0.02304122,0.0001829274,0.0001130706,0.000976612,0.0003524799,0.0001212034,0.00001000559,0.001145924,0.00002128399],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002162104,"about_ca_system_score_gemma":0.1191688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006190588,"about_ca_topic_score_gemma":0.0008731219,"domain_scores_codex":[0.9974203,0.00009941778,0.0006145027,0.0002935845,0.0009453455,0.0006268615],"domain_scores_gemma":[0.9946309,0.0001128044,0.0002617037,0.0003208119,0.001904006,0.00276977],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000425605,0.0007394786,0.003912426,0.0001473345,0.00007422063,0.00001456365,0.0008521327,0.000004884243,0.00002320521,0.002771701,0.3190826,0.6723349],"study_design_scores_gemma":[0.001441864,0.0007116422,0.01199508,0.004216608,0.0002111711,0.0007185836,0.002201295,0.03002265,0.0001029729,0.003472557,0.9444396,0.0004659597],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.3568183,0.0001510288,0.2933533,0.3192562,0.02102643,0.002864856,0.00001054747,0.0001743737,0.006344934],"genre_scores_gemma":[0.2670235,0.00000981352,0.6269063,0.09400587,0.01136578,0.0002118736,0.0001080028,0.00005418603,0.0003147505],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6718689,"threshold_uncertainty_score":0.9941143,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04517756488486576,"score_gpt":0.4255364163158259,"score_spread":0.3803588514309602,"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."}}