{"id":"W4388073046","doi":"10.5070/sr33162445","title":"[SoK] Evaluations in Industrial Intrusion Detection Research","year":2023,"lang":"en","type":"article","venue":"Journal of Systems Research","topic":"Network Security and Intrusion Detection","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Deutsche Forschungsgemeinschaft","keywords":"Benchmarking; Intrusion detection system; Field (mathematics); Data science; Computer science; Risk analysis (engineering); Computer security; Business; Marketing","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.04251798,0.00008600452,0.0002357183,0.004217896,0.0005486492,0.00047595,0.00102458,0.0002380187,0.0000221501],"category_scores_gemma":[0.002439974,0.00007387371,0.00007415866,0.008994084,0.0001005724,0.0009216575,0.0004786279,0.002445465,0.0002896544],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005955371,"about_ca_system_score_gemma":0.0005218344,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00040696,"about_ca_topic_score_gemma":0.0001430494,"domain_scores_codex":[0.9910273,0.003790894,0.0008144502,0.0002678621,0.003460197,0.0006393242],"domain_scores_gemma":[0.9957702,0.001305303,0.0001861833,0.0004064429,0.002152063,0.00017984],"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":[0.0006079514,0.0004781315,0.001917336,0.0001820153,0.00008234797,0.0006625809,0.007764731,0.01678555,0.08309884,0.0151679,0.07956162,0.793691],"study_design_scores_gemma":[0.00416267,0.004214238,0.01119521,0.001950865,0.000007110901,0.0007379774,0.004655618,0.8042573,0.01523373,0.02327476,0.1298684,0.0004420828],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9818945,0.0005426184,0.008444536,0.002267977,0.004587164,0.0007721212,0.000001031492,0.00007113106,0.0014189],"genre_scores_gemma":[0.9975879,0.0003471511,0.00007074395,0.000004421354,0.001515745,0.0000306956,4.85011e-7,0.00001010882,0.0004327288],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7932489,"threshold_uncertainty_score":0.9998559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3379895569946412,"score_gpt":0.4654608749629295,"score_spread":0.1274713179682883,"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."}}