{"id":"W4415380101","doi":"10.1186/s41077-025-00375-x","title":"Using Kane’s validity framework to examine the implications of feedback in simulation-based assessments","year":2025,"lang":"en","type":"article","venue":"Advances in Simulation","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Sinai Health System; University of Toronto; University Health Network; St. Michael's Hospital","funders":"","keywords":"Health services research; Argument (complex analysis); External validity; Predictive validity; Program evaluation; Test validity; Validity; Internal validity","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005274885,0.0001725526,0.0003148692,0.000554669,0.0001153036,0.00001627567,0.000156027,0.0001500795,0.00007276085],"category_scores_gemma":[0.001421709,0.0001600153,0.00005790727,0.00223627,0.00006444882,0.000257414,0.00002858078,0.0002732531,0.000004908817],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000667687,"about_ca_system_score_gemma":0.0004842377,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001582287,"about_ca_topic_score_gemma":0.0001151096,"domain_scores_codex":[0.9980036,0.000198651,0.000855232,0.0003724133,0.0003190881,0.0002510231],"domain_scores_gemma":[0.9919896,0.006477191,0.0002532942,0.0006646848,0.0005443391,0.00007088023],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.00007557822,0.0000965216,0.4650089,0.0000834594,0.000003266115,1.819884e-7,0.0001741548,0.5311996,0.00004803661,0.0004724558,0.000001587196,0.002836324],"study_design_scores_gemma":[0.0005366628,0.00003994474,0.5073476,0.0003992716,0.00001906819,8.39081e-8,0.0001773895,0.4875535,0.0001302683,0.003466019,0.0002552851,0.00007480889],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7400473,0.0002653794,0.2556061,0.001824837,0.0004015747,0.001446395,0.00001418173,0.00003556016,0.0003586941],"genre_scores_gemma":[0.9842356,0.000009986171,0.01427702,0.001172386,0.00008962981,0.00009149257,0.00007355805,0.0000196452,0.00003067143],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2441883,"threshold_uncertainty_score":0.6525236,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1389926103905232,"score_gpt":0.527348115375092,"score_spread":0.3883555049845687,"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."}}