{"id":"W2169719505","doi":"10.1109/icsme.2014.26","title":"Why Do Automated Builds Break? An Empirical Study","year":2014,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":74,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Leverage (statistics); Executable; Computer science; Software engineering; Breakage; Software; Empirical research; Work (physics); Engineering; World Wide Web; Operating system; Artificial intelligence","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01467221,0.0005769539,0.00045821,0.005072307,0.001992831,0.003202739,0.002474464,0.001882616,0.00313218],"category_scores_gemma":[0.1381857,0.001050635,0.0004195705,0.005089153,0.002260077,0.004953699,0.001911656,0.002619874,0.000825829],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001811099,"about_ca_system_score_gemma":0.00168064,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004189972,"about_ca_topic_score_gemma":0.006217917,"domain_scores_codex":[0.9812899,0.008202381,0.001899723,0.001971231,0.00512104,0.001515744],"domain_scores_gemma":[0.5451316,0.3295613,0.08580475,0.01375829,0.02104585,0.004698182],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002675648,0.001290457,0.9303147,0.0002968954,0.0001079229,0.00156406,0.02887608,0.0009806668,0.000577242,0.001026531,0.002160062,0.03253788],"study_design_scores_gemma":[0.00004114352,0.0006577623,0.9231393,0.0002565085,0.00008302505,0.002056275,0.05720237,0.005334937,0.0008174026,0.0009308361,0.009411324,0.0000690145],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9972937,0.0002228915,0.0006035341,0.0002860561,0.00000777487,0.00007750119,0.0001748211,0.0000247828,0.001309033],"genre_scores_gemma":[0.9982804,0.0002348019,0.0007053365,0.0000915057,0.00001638024,0.00006834781,0.0002219649,0.00002263685,0.0003585281],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9853278,"threshold_uncertainty_score":0.07759494,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02278950590193803,"score_gpt":0.3377818139732754,"score_spread":0.3149923080713374,"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."}}