{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001333527,0.0005781457,0.0003649289,0.002083298,0.000272307,0.0007333471,0.000671218,0.0007794251,0.001956264],"category_scores_gemma":[0.0120717,0.0002675749,0.0003689493,0.0008456421,0.0003895025,0.001152041,0.0004604485,0.0007121507,0.0007915787],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007603441,"about_ca_system_score_gemma":0.0004087143,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008603476,"about_ca_topic_score_gemma":0.01271545,"domain_scores_codex":[0.9992446,0.000086452,0.00004141154,0.0001774431,0.000310271,0.0001397666],"domain_scores_gemma":[0.9915873,0.005103894,0.0009080413,0.000471206,0.001467464,0.0004620123],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001889486,0.0003399068,0.383361,0.0003255422,0.0001571955,0.001287931,0.0003646171,0.4214962,0.03575298,0.003416033,0.007917745,0.1436914],"study_design_scores_gemma":[0.00003249168,0.000354501,0.08154654,0.00006449276,0.00006399935,0.0004597116,0.0001843738,0.8850352,0.02621403,0.003213539,0.00276744,0.00006372377],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9693367,0.0008703384,0.02245865,0.0002326604,0.00003838855,0.00003148173,0.002000591,0.001540721,0.00349052],"genre_scores_gemma":[0.9900664,0.000145484,0.006690759,0.00004006651,0.000008146265,0.00001256132,0.001748881,0.00013247,0.001155218],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008603476,"threshold_uncertainty_score":0.01710683,"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."}}