{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001430768,0.002254255,0.0008523336,0.002527674,0.0003095401,0.001123035,0.001088591,0.0007898475,0.001096942],"category_scores_gemma":[0.01244583,0.0004149337,0.000441846,0.0008013169,0.0002488092,0.001315541,0.000997475,0.0006998209,0.001292393],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003373202,"about_ca_system_score_gemma":0.0004063619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003067176,"about_ca_topic_score_gemma":0.004525022,"domain_scores_codex":[0.9981899,0.0002404611,0.000119207,0.0005634887,0.0007280867,0.0001588496],"domain_scores_gemma":[0.9931881,0.002816793,0.001390837,0.0009862453,0.001378883,0.0002391772],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002718719,0.0007144858,0.1924548,0.001267459,0.0004556459,0.0007102375,0.001332619,0.006642278,0.1181444,0.001102689,0.02694392,0.6475127],"study_design_scores_gemma":[0.0001856185,0.003194001,0.3968208,0.0003064842,0.0004350128,0.002043858,0.0007788932,0.3882581,0.1815394,0.002464648,0.02360826,0.0003649165],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7325167,0.002814326,0.1605328,0.0002064071,0.0002032383,0.0004892764,0.003865209,0.09542102,0.003950996],"genre_scores_gemma":[0.9388454,0.0003082042,0.05475215,0.000199377,0.00004674313,0.0001843412,0.002336121,0.001399298,0.001928356],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003067176,"threshold_uncertainty_score":0.00756669,"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."}}