{"id":"W6884736400","doi":"10.1192/j.eurpsy.2023.919","title":"Criminal responsibility evaluations: Benchmarking in different countries","year":2023,"lang":"en","type":"article","venue":"PubMed Central","topic":"Psychopathy, Forensic Psychiatry, Sexual Offending","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Criminal responsibility; Sanity; Mental health; Insanity; Agency (philosophy); Benchmarking; PsycINFO; Meaning (existential)","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":[0.02941781,0.0004550451,0.001067854,0.01878519,0.0006433863,0.003456556,0.001301226,0.00110236,0.001995018],"category_scores_gemma":[0.08628042,0.0003910834,0.001536352,0.0217667,0.001489966,0.002909657,0.003450012,0.0008205887,0.0005217949],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003180235,"about_ca_system_score_gemma":0.003777451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005920569,"about_ca_topic_score_gemma":0.005618435,"domain_scores_codex":[0.9631789,0.01951934,0.007873552,0.002666094,0.005289641,0.001472475],"domain_scores_gemma":[0.8992711,0.0488668,0.0171541,0.004919067,0.02699205,0.002796939],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002111521,0.0009501736,0.7032622,0.0126827,0.002152266,0.001300013,0.01363447,0.004853927,0.0004815673,0.00390188,0.01112732,0.243542],"study_design_scores_gemma":[0.0001331658,0.0006825401,0.9606624,0.004870213,0.0004551801,0.001308957,0.01580724,0.001098212,0.0005713394,0.0005763386,0.01375781,0.00007657264],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9347664,0.03240755,0.002001955,0.001350844,0.0001327197,0.001207873,0.008923464,0.0001619828,0.01904736],"genre_scores_gemma":[0.9814023,0.008831099,0.002876723,0.0002620184,0.00003470936,0.0007039896,0.005306556,0.00007123373,0.0005114025],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02941781,"threshold_uncertainty_score":0.155578,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05797350371796232,"score_gpt":0.3511283806649363,"score_spread":0.293154876946974,"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."}}