{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005564464,0.0000566484,0.00005910478,0.0002929719,0.00013497,0.00008274306,0.000139172,0.00003425939,0.000002099025],"category_scores_gemma":[0.001201528,0.00005441947,0.00000667937,0.00073695,0.00008080115,0.0001293955,0.0001289515,0.0001293486,0.000003413817],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001497584,"about_ca_system_score_gemma":0.00006818413,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003927464,"about_ca_topic_score_gemma":0.00004221126,"domain_scores_codex":[0.9991962,0.00006185924,0.00009238369,0.0003654749,0.0001426436,0.0001414365],"domain_scores_gemma":[0.9992479,0.0003363347,0.00001394807,0.0003046961,0.00007178845,0.00002534532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00004321735,0.002630737,0.6280366,0.0001923679,0.00002524637,0.002494766,0.00539489,0.0002400279,0.002316795,0.1723269,0.05806443,0.128234],"study_design_scores_gemma":[0.001352874,0.0008638522,0.06142004,0.0005881501,0.000008098594,0.0008720213,0.0005691275,0.2012192,0.02765061,0.7028661,0.001941292,0.0006485729],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06465808,0.0003640297,0.9085438,0.01358286,0.0001788383,0.000907655,0.000001457316,0.004584455,0.007178837],"genre_scores_gemma":[0.8970857,0.000006429442,0.1019394,0.00009540503,0.00001495082,0.00003767948,3.768884e-7,0.000003000445,0.0008171216],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8324276,"threshold_uncertainty_score":0.2219162,"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."}}