{"id":"W4283080616","doi":"10.1109/icse-seip55303.2022.9793941","title":"The Impact of Flaky Tests on Historical Test Prioritization on Chrome","year":2022,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Prioritization; Blocking (statistics); Pipeline (software); Computer science; Test (biology); Reliability engineering; Replication (statistics); Work (physics); Regression testing; Software; Engineering; Software development; Operating system; Computer network","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":[0.0003275938,0.00008052109,0.00008759843,0.0000787905,0.0003023682,0.00004007622,0.0005991011,0.00001751548,0.00001180687],"category_scores_gemma":[0.0007920041,0.00005157735,0.00006159313,0.0004187658,0.00001696718,0.00005117848,0.0001782596,0.000137742,0.000006554196],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005900372,"about_ca_system_score_gemma":0.00008472329,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002312938,"about_ca_topic_score_gemma":0.000001267664,"domain_scores_codex":[0.999153,0.0000548874,0.0001439694,0.0001886611,0.0003156196,0.0001438171],"domain_scores_gemma":[0.9979118,0.001423698,0.00007876329,0.0005002368,0.00004917519,0.00003635602],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00003120535,0.001054113,0.05365568,0.000009334728,0.00002002429,0.0000268137,0.0004901075,0.002212751,0.001400138,0.02622869,0.8001169,0.1147542],"study_design_scores_gemma":[0.0011397,0.02641934,0.5193407,0.0001242725,0.00001450723,0.0001994019,0.0000115802,0.2337,0.005238699,0.1923454,0.02026996,0.001196385],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.366468,0.0007226608,0.4700534,0.01530252,0.003641755,0.002313869,0.00003670362,0.06805298,0.07340809],"genre_scores_gemma":[0.9922662,0.000002405631,0.006913432,0.0001259918,0.00003117245,0.00002239336,0.000001038665,0.000006838227,0.0006304906],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.779847,"threshold_uncertainty_score":0.2325602,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01932153530382496,"score_gpt":0.2868846522061037,"score_spread":0.2675631169022787,"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."}}