{"id":"W4400484638","doi":"10.1145/3663529.3663780","title":"AutoOffAB: Toward Automated Offline A/B Testing for Data-Driven Requirement Engineering","year":2024,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia, Okanagan Campus; Kelowna General Hospital; University of British Columbia","funders":"","keywords":"Computer science; Software engineering; Reliability engineering; Engineering","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":[0.005741012,0.00204625,0.0008217237,0.003040165,0.0005322126,0.002756239,0.003377174,0.001250109,0.004160813],"category_scores_gemma":[0.02598786,0.001242426,0.00119235,0.0008693206,0.001344142,0.003868237,0.003835179,0.002634115,0.003351886],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006639821,"about_ca_system_score_gemma":0.002083093,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003680671,"about_ca_topic_score_gemma":0.003930162,"domain_scores_codex":[0.9920874,0.002952265,0.0004666066,0.001213591,0.002870582,0.0004096235],"domain_scores_gemma":[0.9752758,0.01050242,0.002124856,0.007847875,0.003373678,0.0008755029],"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":[0.00110534,0.001257572,0.02056973,0.0009413847,0.0002161298,0.0007632885,0.001555228,0.03795286,0.06458452,0.01818059,0.03473955,0.8181339],"study_design_scores_gemma":[0.000214271,0.0007195704,0.006599548,0.0004118005,0.00007474453,0.001156876,0.0005516435,0.8338977,0.07243744,0.03413987,0.04959293,0.0002036126],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01646355,0.0002038519,0.9073421,0.000298577,0.00006478645,0.0005214431,0.0004719917,0.07156803,0.003065581],"genre_scores_gemma":[0.1847489,0.0001577999,0.8059264,0.000365055,0.00003094817,0.0004181862,0.0017532,0.004254195,0.002345358],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005741012,"threshold_uncertainty_score":0.03036177,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1244871187603879,"score_gpt":0.3383702706052155,"score_spread":0.2138831518448276,"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."}}