{"id":"W4413048668","doi":"10.1080/10826084.2025.2544301","title":"Performance Validity Testing in Patients with Substance Abuse in Addiction Care","year":2025,"lang":"en","type":"article","venue":"Substance Use & Misuse","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Addiction; Substance abuse; Psychiatry; Psychology; Substance use; Test validity; Substance Abuse Detection; Clinical psychology; Psychometrics; Medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0001399733,0.0003797007,0.0005437761,0.0004562038,0.0001163692,0.00003450651,0.0001497957,0.0001443355,0.00002625547],"category_scores_gemma":[0.0000856244,0.0003232484,0.00006578829,0.001966907,0.00009332472,0.0006820008,0.000007445637,0.0003870242,0.00001997722],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008965287,"about_ca_system_score_gemma":0.0001336524,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006583734,"about_ca_topic_score_gemma":0.008239558,"domain_scores_codex":[0.9978065,0.00005941528,0.0005057667,0.0006020482,0.0004203795,0.0006059104],"domain_scores_gemma":[0.9985388,0.0002507119,0.0001394001,0.0006263892,0.0003451954,0.00009953161],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001690866,0.0003388538,0.9889053,0.00009328268,0.000034434,0.00008896382,0.00614462,0.0001175021,0.00004229842,0.00001165833,0.00008795659,0.002444253],"study_design_scores_gemma":[0.01123927,0.0002100769,0.9838598,0.002195465,0.00008462554,0.000001371782,0.0007653628,0.0001309859,0.0009924467,0.000007597842,0.0002236102,0.0002893432],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9969602,0.0009558064,0.00001076316,0.00003080547,0.0002205076,0.0009940449,0.000008388991,0.0001608372,0.0006586868],"genre_scores_gemma":[0.9973767,0.0004482908,0.001271724,0.0001735015,0.00003007041,0.00009904204,0.00009484258,0.00004378762,0.0004620372],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009548401,"threshold_uncertainty_score":0.999922,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03169179198422639,"score_gpt":0.2594610720003583,"score_spread":0.2277692800161319,"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."}}