{"id":"W4414121223","doi":"10.1080/09546553.2025.2555221","title":"How Rigorous are Evaluations of Violent Extremism Prevention Programs? Results from a Systematic Methodological Review","year":2025,"lang":"en","type":"article","venue":"Terrorism and Political Violence","topic":"Terrorism, Counterterrorism, and Political Violence","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Montréal; Université de Sherbrooke; Université du Québec à Trois-Rivières","funders":"","keywords":"Violent extremism; Terrorism; Poison control; Human factors and ergonomics; Suicide prevention","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002960807,0.0003478526,0.001057285,0.0001257158,0.0003361601,0.0001257407,0.0005836795,0.0003364813,0.00004068946],"category_scores_gemma":[0.005725318,0.0002804718,0.0002629719,0.0004140548,0.0009543991,0.000283938,0.0001556737,0.0003693821,0.000014888],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001686661,"about_ca_system_score_gemma":0.0002745378,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002057415,"about_ca_topic_score_gemma":0.0001818146,"domain_scores_codex":[0.9939269,0.002500505,0.001116289,0.0007367615,0.0007915919,0.0009279124],"domain_scores_gemma":[0.996001,0.002050902,0.0004352261,0.0006724203,0.0003616667,0.0004787632],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0006918173,0.003702912,0.01348515,0.09644736,0.0009227624,0.0001221987,0.05159906,0.000001627758,0.0006142941,0.5375549,0.003441303,0.2914167],"study_design_scores_gemma":[0.002845716,0.0007626271,0.02405096,0.445828,0.002648599,0.00001833805,0.183774,0.0004956743,0.0005416008,0.3345889,0.002818503,0.001627109],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7804341,0.0941992,0.04593549,0.05730658,0.002815071,0.00881138,0.0007921941,0.0007776489,0.008928276],"genre_scores_gemma":[0.9908054,0.00625609,0.0004758215,0.001237013,0.0002214996,0.0002407538,0.0000407929,0.0000128917,0.0007097003],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3493806,"threshold_uncertainty_score":0.9999648,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1198659355755358,"score_gpt":0.4232939892298018,"score_spread":0.3034280536542661,"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."}}