{"id":"W2560768181","doi":"10.1186/s41077-016-0033-y","title":"Validation of educational assessments: a primer for simulation and beyond","year":2016,"lang":"en","type":"article","venue":"Advances in Simulation","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":377,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Health services research; Primer (cosmetics); Public health; Medical education; Medicine; Psychology; Nursing; Chemistry","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":[0.0002497825,0.000111686,0.0001912091,0.0002458222,0.00005196698,0.000006558138,0.0000344957,0.0000852439,0.0001248235],"category_scores_gemma":[0.0005749643,0.00009339986,0.00003648522,0.0002377369,0.00005800439,0.0006878332,0.000009030432,0.00004334672,0.000003398677],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002032379,"about_ca_system_score_gemma":0.0001936375,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007429241,"about_ca_topic_score_gemma":0.000002708592,"domain_scores_codex":[0.9987957,0.00004849985,0.0005088418,0.0002713767,0.0002326634,0.000142956],"domain_scores_gemma":[0.995607,0.003304599,0.000263889,0.0001930508,0.0005699184,0.0000615747],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002587612,0.0001157052,0.6629334,0.000181045,0.000008918395,4.315803e-8,0.000263194,0.3050016,0.0007166807,0.002597207,0.000005220039,0.02791822],"study_design_scores_gemma":[0.005385809,0.0004689822,0.6201816,0.0005231939,0.00008001888,0.000001685763,0.0002884688,0.3318868,0.004414129,0.03453877,0.001938811,0.0002917374],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9267001,0.0006215377,0.06870894,0.001893419,0.0003634174,0.001285202,0.00002780436,0.00002232904,0.0003772868],"genre_scores_gemma":[0.9910643,0.00003910437,0.008192815,0.0001244944,0.0001692192,0.0001053796,0.0001384041,0.00001928231,0.0001470167],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06436422,"threshold_uncertainty_score":0.3808736,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03932502403537743,"score_gpt":0.4675536437142719,"score_spread":0.4282286196788945,"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."}}