{"id":"W4401420422","doi":"10.1145/3643656.3643899","title":"Predicting the Lifetime of Flaky Tests on Chrome","year":2024,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Materials science; Reliability engineering; Computer science; Engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.0003619275,0.00006843424,0.00006882948,0.00006649616,0.00005420059,0.0001033555,0.0005450018,0.00002700754,0.000009939197],"category_scores_gemma":[0.0002829414,0.00003919921,0.00003915415,0.0003434847,0.00003432183,0.00009702702,0.0001519892,0.0001100932,0.00004752566],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001253597,"about_ca_system_score_gemma":0.00004232204,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006031442,"about_ca_topic_score_gemma":0.000001100405,"domain_scores_codex":[0.9993542,0.00002444403,0.0001245523,0.0001918326,0.000185681,0.0001193073],"domain_scores_gemma":[0.9985584,0.0009326587,0.0000230086,0.0004327447,0.00002824023,0.00002492212],"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":[0.000005835426,0.0001679474,0.02596852,0.0002475338,0.00008690342,0.0001029499,0.004198062,0.0002188449,0.002417727,0.2054878,0.3846594,0.3764384],"study_design_scores_gemma":[0.0001286467,0.0008200004,0.02782119,0.001227998,0.0000182662,0.0001478553,0.00001039408,0.7733008,0.02096253,0.1676811,0.007477054,0.0004041635],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1044802,0.001082476,0.7235981,0.009873088,0.001486255,0.0004579833,0.000004565573,0.0880594,0.07095794],"genre_scores_gemma":[0.9582924,0.000002854255,0.0408514,0.0003051289,0.00007434401,0.000007957429,2.354563e-7,0.000006451497,0.0004592411],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8538122,"threshold_uncertainty_score":0.1598497,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01636937012247894,"score_gpt":0.2733729501834212,"score_spread":0.2570035800609423,"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."}}