{"id":"W4411691791","doi":"10.1002/aet2.70063","title":"Development and Initial Validity Evidence for the EvaLeR Tool: Assessing Quality of Emergency Medicine Educational Resources","year":2025,"lang":"en","type":"article","venue":"AEM Education and Training","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"Society for Academic Emergency Medicine","keywords":"Cronbach's alpha; Intraclass correlation; Reliability (semiconductor); Quality (philosophy); Resource (disambiguation); Content validity; Psychology; Medical education; Computer science; Medicine; Psychometrics; Clinical psychology","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":[0.002057762,0.00008980143,0.0001737268,0.0001418019,0.0002991637,0.00001466502,0.00006345436,0.00005377127,0.0001674356],"category_scores_gemma":[0.006670235,0.00006634348,0.00002264098,0.0003497181,0.0001676265,0.0001295105,0.00001991288,0.0001218394,4.196032e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005285321,"about_ca_system_score_gemma":0.001348879,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000340089,"about_ca_topic_score_gemma":0.000006231351,"domain_scores_codex":[0.9988253,0.00005918031,0.0005809863,0.0001965064,0.0002277282,0.000110344],"domain_scores_gemma":[0.9981747,0.0009332792,0.0002040704,0.0001609188,0.0004928205,0.00003425928],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00006854255,0.0002609142,0.04990004,0.0009082401,0.00008874099,2.600041e-8,0.03114444,6.15192e-7,0.0008783534,0.03567643,0.01017496,0.8708987],"study_design_scores_gemma":[0.0004379278,0.00003463891,0.9412381,0.0008092126,0.0001193656,0.000004318452,0.02015273,0.0001285967,0.0001894872,0.002501567,0.03430006,0.00008398472],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.960388,0.001386411,0.002849968,0.03103394,0.001782865,0.0003804804,7.752673e-7,0.00001216764,0.002165394],"genre_scores_gemma":[0.9824362,0.00009511818,0.01455113,0.001060942,0.0004987889,0.0001546658,0.00003153203,0.00000629733,0.001165346],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8913381,"threshold_uncertainty_score":0.7985376,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4127543549796017,"score_gpt":0.5511999300413459,"score_spread":0.1384455750617442,"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."}}