{"id":"W2910794638","doi":"10.1145/3287624.3287661","title":"Suspect2vec","year":2019,"lang":"en","type":"article","venue":"","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Debugging; Computer science; Suspect; Boolean satisfiability problem; Overhead (engineering); Algorithmic program debugging; Key (lock); Baseline (sea); Prioritization; Solver; Satisfiability; Software deployment; Programming language; Algorithm; Software engineering; Operating system","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":[0.0009273545,0.002049285,0.0008117039,0.001363396,0.0005663119,0.001861946,0.00284242,0.001477938,0.02903091],"category_scores_gemma":[0.005556548,0.0008984416,0.001532766,0.001018237,0.0004814145,0.002429812,0.001483453,0.001478246,0.01401262],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001002486,"about_ca_system_score_gemma":0.001748101,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008063799,"about_ca_topic_score_gemma":0.01983511,"domain_scores_codex":[0.9991164,0.0001640901,0.0000530984,0.0002909581,0.0003057896,0.00006968851],"domain_scores_gemma":[0.9983073,0.0007078623,0.00008696187,0.0004934592,0.000357906,0.00004643746],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001075189,0.0002415632,0.009881547,0.001229268,0.0002348083,0.0003395076,0.0002397103,0.05040843,0.006657303,0.01802849,0.6498995,0.2617646],"study_design_scores_gemma":[0.0002561605,0.0002887772,0.001773578,0.0001332212,0.00009205338,0.0004161418,0.0001488797,0.7011715,0.02110554,0.03381098,0.2407185,0.00008469025],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"software","genre_gemma":"empirical","genre_scores_codex":[0.03186092,0.001833322,0.3337542,0.001784998,0.001121011,0.000544915,0.09818553,0.5037875,0.02712766],"genre_scores_gemma":[0.2269629,0.001249417,0.3345284,0.001854627,0.0002284797,0.0009582314,0.3710487,0.02649477,0.03667445],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02903091,"threshold_uncertainty_score":0.09711808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00897527818925105,"score_gpt":0.2492333524332143,"score_spread":0.2402580742439633,"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."}}