{"id":"W7125599361","doi":"10.1109/cascon66301.2025.00115","title":"LLM-Driven Event Log Generation from Forensic Cases: A Comparative Study of ChatGPT, Claude, and Gemini","year":2025,"lang":"","type":"article","venue":"","topic":"Business Process Modeling and Analysis","field":"Business, Management and Accounting","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Timeline; Digital forensics; Schema (genetic algorithms); Bespoke; Event (particle physics); Context (archaeology); Consistency (knowledge bases); Visualization","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":[0.01951015,0.0006164825,0.0005371466,0.002926079,0.0008834603,0.002545218,0.002210645,0.001063367,0.001947343],"category_scores_gemma":[0.07588898,0.0005763532,0.0006073753,0.001888795,0.00143315,0.004893463,0.003316101,0.001846494,0.0008231139],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001764582,"about_ca_system_score_gemma":0.002147549,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004214992,"about_ca_topic_score_gemma":0.005212784,"domain_scores_codex":[0.9907022,0.005711447,0.0005677916,0.0007941902,0.002011264,0.0002131368],"domain_scores_gemma":[0.9395694,0.04535535,0.001510591,0.008578241,0.004377624,0.000608763],"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":[0.002263921,0.002000803,0.04958644,0.002713471,0.0003926376,0.001775818,0.03596763,0.09199393,0.02112933,0.03813118,0.01978274,0.7342622],"study_design_scores_gemma":[0.0003218017,0.001619159,0.02672032,0.0006797861,0.0002377,0.002514016,0.01377085,0.8005621,0.06037503,0.02091415,0.07194206,0.000343074],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6914116,0.001236967,0.2697915,0.002698586,0.0001170393,0.001339968,0.002457868,0.01555807,0.01538839],"genre_scores_gemma":[0.7013451,0.0004880905,0.2898562,0.0003187693,0.00003842768,0.0003863056,0.004097449,0.001223279,0.00224639],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01951015,"threshold_uncertainty_score":0.1031808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0484112145332152,"score_gpt":0.2955463398487193,"score_spread":0.2471351253155041,"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."}}