{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003962829,0.0003696851,0.0004515258,0.002011635,0.0007514309,0.001063882,0.0005821965,0.0004817868,0.001379271],"category_scores_gemma":[0.016558,0.0002116568,0.00050169,0.001716878,0.0008135572,0.0005684591,0.001206282,0.0006677,0.000333563],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006959133,"about_ca_system_score_gemma":0.001047572,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006842339,"about_ca_topic_score_gemma":0.007460823,"domain_scores_codex":[0.9976801,0.0007045863,0.0004755305,0.000206914,0.0006047967,0.000328083],"domain_scores_gemma":[0.9902919,0.002641923,0.004601603,0.0003824326,0.001513631,0.0005685197],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001707351,0.00001207702,0.9980672,0.000008716072,0.000007711188,0.00007554482,0.0001038235,0.00002299342,0.00003285464,0.00001213112,0.00005314831,0.001586769],"study_design_scores_gemma":[0.000003537149,0.0001480941,0.9969399,0.0000541143,0.00002391092,0.0008927881,0.0006148076,0.0005788282,0.0002945148,0.0001244917,0.0003192484,0.000005813753],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9981728,0.0004002538,0.0002042519,0.0001015079,0.00000995873,0.0000238616,0.0001378312,0.000007481683,0.0009420548],"genre_scores_gemma":[0.9994018,0.00009602896,0.0002119999,0.00003813486,0.000006299802,0.00001283538,0.0001418354,0.000002839849,0.00008822153],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006842339,"threshold_uncertainty_score":0.02095771,"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."}}