{"id":"W7125894626","doi":"10.1109/ase63991.2025.00308","title":"MobileUPReg: Identifying User-Perceived Performance Regressions in Mobile OS Versions","year":2025,"lang":"","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; Concordia University","funders":"","keywords":"Wilcoxon signed-rank test; Regression; Mobile device; Focus (optics); Regression analysis; Regression testing; Baseline (sea); Linear regression","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0008688452,0.0005023835,0.0005614462,0.001040237,0.0008645401,0.0005273661,0.002285622,0.0003453462,0.0002220075],"category_scores_gemma":[0.0005877481,0.0004847788,0.0002040178,0.002962004,0.0002394785,0.001147735,0.001855599,0.0008662946,0.0001769997],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003453775,"about_ca_system_score_gemma":0.0005659808,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004341549,"about_ca_topic_score_gemma":0.00005010043,"domain_scores_codex":[0.996085,0.0002574446,0.0009128153,0.001278927,0.0005021463,0.0009636619],"domain_scores_gemma":[0.9963685,0.000997834,0.0002000562,0.001980047,0.0002448525,0.0002087284],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00005933085,0.001127325,0.4709436,0.0006258369,0.00007819061,0.0001286151,0.005036134,0.002626453,0.001256617,0.002839857,0.1432376,0.3720404],"study_design_scores_gemma":[0.002145507,0.0006123578,0.348407,0.01639087,0.00008007618,0.00005386096,0.0004337384,0.6038755,0.00573496,0.009642025,0.01082296,0.00180108],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7543707,0.002287167,0.2073674,0.0007511894,0.002990994,0.001740577,0.0000098503,0.01354936,0.01693272],"genre_scores_gemma":[0.9223691,0.0009507353,0.06557115,0.0005325551,0.0000432863,0.0003172483,0.00000374999,0.00002435077,0.01018779],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.601249,"threshold_uncertainty_score":0.9997604,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02701462286612364,"score_gpt":0.3113889806829244,"score_spread":0.2843743578168008,"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."}}