{"id":"W2973296283","doi":"10.1109/tse.2019.2941880","title":"Studying the Impact of Noises in Build Breakage Data","year":2019,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Computer science; Replicate; Breakage; Server; Troubleshooting; Timeout; Data mining; Data science; World Wide Web; Statistics; 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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000491977,0.0002157279,0.000250957,0.0004031496,0.00004797727,0.00007529557,0.002000687,0.00007052805,0.00003313011],"category_scores_gemma":[0.0001610657,0.0001698318,0.0001116845,0.001103512,0.00002004415,0.0006065948,0.00003255279,0.0005302523,0.0000436043],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001490088,"about_ca_system_score_gemma":0.00009597482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002672116,"about_ca_topic_score_gemma":0.000008892246,"domain_scores_codex":[0.9984145,0.00003407052,0.0002881255,0.0004423622,0.0004142564,0.0004066722],"domain_scores_gemma":[0.9963319,0.001607275,0.00003965042,0.001891791,0.00005187793,0.000077561],"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":[0.000007315232,0.00009291428,0.009051546,0.00004812876,0.00006435036,0.000008921363,0.0003189974,0.9827649,0.001136679,0.00001997978,0.00004456496,0.006441706],"study_design_scores_gemma":[0.001169737,0.0003452893,0.1619028,0.0003780541,0.00001653023,0.00005294692,0.0000449868,0.8256409,0.00962032,0.00001518156,0.0001639772,0.0006492631],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3133706,0.00006759117,0.6856233,0.00001527909,0.0003842734,0.0002328086,0.00002279856,0.0002781632,0.000005114007],"genre_scores_gemma":[0.984499,0.00001612935,0.01534827,0.000006354636,0.00002261666,0.00002576647,0.000001840384,0.00003416432,0.00004591166],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6711283,"threshold_uncertainty_score":0.6925538,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02816949054392912,"score_gpt":0.2874072299796261,"score_spread":0.2592377394356969,"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."}}