{"id":"W4410538223","doi":"10.1109/icst62969.2025.10989029","title":"ML-Based Test Case Prioritization: A Research and Production Perspective in CI Environments","year":2025,"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":"Ontario Tech University","funders":"","keywords":"Prioritization; Perspective (graphical); Test (biology); Production (economics); Computer science; Reliability engineering; Risk analysis (engineering); Engineering; Process management; Artificial intelligence; Business; Geology","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":[0.02683988,0.002048988,0.001050899,0.003382745,0.0007723317,0.005539024,0.004943778,0.001753115,0.001557372],"category_scores_gemma":[0.08950608,0.0007860171,0.0006259084,0.002892208,0.002226126,0.008872403,0.002878406,0.004887254,0.0006228791],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003086037,"about_ca_system_score_gemma":0.005622246,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008792123,"about_ca_topic_score_gemma":0.009651452,"domain_scores_codex":[0.9858094,0.007052373,0.001076053,0.001963696,0.00324413,0.0008543115],"domain_scores_gemma":[0.9260203,0.0456747,0.004965107,0.01111987,0.01006676,0.002153208],"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":[0.0006845194,0.001439039,0.05034353,0.0007270304,0.0001481609,0.0004594336,0.001268522,0.3639475,0.02438633,0.01520924,0.01447887,0.5269079],"study_design_scores_gemma":[0.00007868704,0.0004139013,0.00516013,0.00012737,0.00004056573,0.0002390029,0.0006138331,0.9607613,0.01465701,0.01247449,0.005376635,0.00005697214],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1717273,0.002291772,0.7943545,0.009111583,0.0002438393,0.0006830971,0.0007401913,0.01475068,0.006097053],"genre_scores_gemma":[0.6732147,0.0004595636,0.3230883,0.0005606557,0.0001245954,0.0002194619,0.00106429,0.0006174899,0.000650964],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02683988,"threshold_uncertainty_score":0.1419446,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04430461546752107,"score_gpt":0.3633840419375879,"score_spread":0.3190794264700668,"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."}}