{"id":"W4400242288","doi":"10.1145/3643991.3644914","title":"A Mutation-Guided Assessment of Acceleration Approaches for Continuous Integration: An Empirical Study of YourBase","year":2024,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal; University of Waterloo","funders":"","keywords":"Acceleration; Computer science; Mutation; Empirical research; Physics; Mathematics; Genetics; Biology; Statistics","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":[0.0005388581,0.00008369765,0.0001608132,0.0001380852,0.00003911081,0.0001238799,0.0002534956,0.00003444442,0.000003052114],"category_scores_gemma":[0.0001411807,0.0000649198,0.00004136076,0.0003006378,0.00001813915,0.0003391214,0.00004593196,0.00005446807,2.170849e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002398915,"about_ca_system_score_gemma":0.0001147485,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008432803,"about_ca_topic_score_gemma":0.00001933865,"domain_scores_codex":[0.9990526,0.00007255756,0.0003330788,0.0002651573,0.000198256,0.00007833297],"domain_scores_gemma":[0.9990366,0.0003606773,0.00007733855,0.0003028661,0.0001936248,0.00002892638],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000393561,0.007336198,0.09983567,0.0005413764,0.0002358966,0.00003709754,0.04482185,0.00235464,0.001825898,0.1505702,0.04460578,0.647796],"study_design_scores_gemma":[0.0002310588,0.001424336,0.00661684,0.00003806416,0.00001737731,0.000008861351,0.0003230062,0.9752208,0.002906638,0.01310713,0.00001352116,0.00009239555],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2329929,0.00001246978,0.7644637,0.0001407865,0.00007081173,0.0004133757,0.000001593004,0.001656792,0.0002475324],"genre_scores_gemma":[0.6568339,1.413447e-7,0.3430021,0.00002056947,0.00001802226,0.00008643624,0.000007021019,0.000004350351,0.00002748018],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9728661,"threshold_uncertainty_score":0.2647353,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1888561377512039,"score_gpt":0.421461520203657,"score_spread":0.2326053824524532,"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."}}