{"id":"W4416754311","doi":"10.1101/2025.11.25.25340922","title":"Synthetic Validation of Pediatric Trust Instruments using Persona-Driven Large Language Models","year":2025,"lang":"","type":"preprint","venue":"medRxiv","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Blueprint; Subject-matter expert; Timeline; Set (abstract data type); Reliability (semiconductor); Model validation; Content validity; Process (computing); Validation test; Flagging","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":[],"consensus_categories":[],"category_scores_codex":[0.05137217,0.0009066836,0.00051345,0.001110135,0.0007480608,0.001866211,0.001652688,0.001068747,0.002750074],"category_scores_gemma":[0.1615932,0.0004733819,0.001294202,0.0009164753,0.001520318,0.001477693,0.003289573,0.001672866,0.0008491465],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002250783,"about_ca_system_score_gemma":0.002389092,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004126944,"about_ca_topic_score_gemma":0.004001749,"domain_scores_codex":[0.9485537,0.0430488,0.002127433,0.002866286,0.002856556,0.0005471988],"domain_scores_gemma":[0.7728019,0.185762,0.007300911,0.01418125,0.01879132,0.001162566],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006131937,0.005394162,0.2665237,0.002194969,0.001464553,0.001341868,0.03817601,0.2953252,0.01731848,0.02158675,0.01911009,0.3254324],"study_design_scores_gemma":[0.001194122,0.002727471,0.1130303,0.0008731483,0.0004001256,0.0005884381,0.00683711,0.8024539,0.02461378,0.02341943,0.02350606,0.0003560962],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8437676,0.0001654249,0.1461849,0.0003517626,0.00009068703,0.002333493,0.002360016,0.0006397505,0.004106423],"genre_scores_gemma":[0.8951048,0.00004730385,0.09645198,0.0002240238,0.00002199372,0.002976872,0.004449314,0.0001081968,0.0006155755],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05137217,"threshold_uncertainty_score":0.2716852,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1930729109072419,"score_gpt":0.411192898759674,"score_spread":0.2181199878524321,"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."}}