{"id":"W4220690511","doi":"10.1007/s10664-021-10109-y","title":"Why secret detection tools are not enough: It’s not just about false positives - An industrial case study","year":2022,"lang":"en","type":"article","venue":"Empirical Software Engineering","topic":"Advanced Malware Detection Techniques","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Victoria","funders":"National Security Agency; North Carolina State University; National Science Foundation","keywords":"False positive paradox; Computer security; Computer science; True positive rate; Data science; Artificial intelligence","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":[0.006775165,0.0006166042,0.0002848404,0.001987442,0.00236689,0.002232132,0.001413378,0.002675635,0.003710683],"category_scores_gemma":[0.03838735,0.000436073,0.000461502,0.001430887,0.002491603,0.003074199,0.001624965,0.002016897,0.0007970011],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001650693,"about_ca_system_score_gemma":0.00146828,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003568536,"about_ca_topic_score_gemma":0.007244912,"domain_scores_codex":[0.9933333,0.003140369,0.0002815986,0.0005165879,0.002238506,0.0004896547],"domain_scores_gemma":[0.9115846,0.07458103,0.00327264,0.00409565,0.005498121,0.0009679065],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001727176,0.01073371,0.4185456,0.001489043,0.0002947653,0.03142344,0.03053616,0.03040556,0.01451314,0.04595721,0.02726324,0.3871109],"study_design_scores_gemma":[0.001626188,0.009975377,0.2663657,0.00268764,0.000942971,0.07504883,0.08027659,0.2422021,0.1020041,0.08453266,0.1338554,0.0004822967],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9552059,0.0007359927,0.02080572,0.003567817,0.00005040589,0.0002708932,0.0002860202,0.0002265573,0.01885064],"genre_scores_gemma":[0.9847152,0.0003096749,0.01160713,0.0003094662,0.00002000829,0.0000602753,0.0001454196,0.00006521082,0.002767654],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006775165,"threshold_uncertainty_score":0.03583097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08079733134382909,"score_gpt":0.315110970198614,"score_spread":0.2343136388547849,"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."}}