{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1985836,0.002498587,0.004021766,0.01282491,0.003706466,0.01577115,0.007626358,0.01465055,0.004180823],"category_scores_gemma":[0.2448261,0.001719064,0.003247216,0.006373564,0.04090507,0.02650058,0.01152794,0.02443992,0.001827739],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01158192,"about_ca_system_score_gemma":0.02417037,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005470115,"about_ca_topic_score_gemma":0.003056687,"domain_scores_codex":[0.81381,0.154343,0.01211217,0.004861027,0.01357115,0.001302642],"domain_scores_gemma":[0.5668091,0.387174,0.006336803,0.01528947,0.0217571,0.002633556],"domain_codex":null,"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":[0.00006959515,0.0001686195,0.001175727,0.004081738,0.0001344076,0.00030755,0.008914567,0.003403621,0.00031611,0.7382722,0.02062801,0.2225278],"study_design_scores_gemma":[0.00004149462,0.0001699575,0.000586369,0.02380253,0.00004683232,0.0004102545,0.002776689,0.004024471,0.0004388475,0.7599681,0.2075966,0.0001378559],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001479531,0.2158986,0.5273484,0.2256353,0.006006627,0.001621,0.0002125244,0.0005383592,0.02125971],"genre_scores_gemma":[0.06309557,0.1579724,0.7261955,0.03657364,0.005439124,0.006186707,0.0003792206,0.0005071685,0.0036506],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1985836,"threshold_uncertainty_score":0.9882891,"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."}}