{"id":"W4226139682","doi":"10.5281/zenodo.6415365","title":"Scalable and Accurate Test Case Prioritization in Continuous Integration Contexts","year":2022,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University; University of Ottawa","funders":"","keywords":"Prioritization; Scalability; Test (biology); Computer science; Biology; Engineering; Database; Ecology; Process management","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":["sts"],"consensus_categories":[],"category_scores_codex":[0.0008692246,0.00008884451,0.0001003504,0.0002460051,0.00164153,0.0007386112,0.0005786678,0.00002569584,0.0003031353],"category_scores_gemma":[0.001412278,0.0001002527,0.00001435393,0.0007147479,0.00005512523,0.0003789523,0.001331461,0.0002312688,0.00008997277],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001298868,"about_ca_system_score_gemma":0.000004732559,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007478494,"about_ca_topic_score_gemma":0.000001121504,"domain_scores_codex":[0.9987634,0.0002896594,0.0001939346,0.0003402941,0.0002065351,0.0002061441],"domain_scores_gemma":[0.9992077,0.0001045143,0.00008408542,0.0003058219,0.0002282928,0.00006953155],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002673764,0.0003111182,0.000292287,0.00003978705,0.000009023207,0.0007483842,0.004491088,0.0001495793,0.002075504,0.01489642,0.12339,0.8535701],"study_design_scores_gemma":[0.002949011,0.00246778,0.005764793,0.0001919228,0.00001967698,0.02200216,0.001008144,0.2440117,0.001752573,0.02366271,0.6949797,0.001189802],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1741203,0.0005121786,0.7609312,0.003268544,0.0003714929,0.002007597,0.0002159943,0.02856511,0.03000757],"genre_scores_gemma":[0.9948009,0.00001312839,0.00452155,0.0001432821,0.00002021567,2.181762e-7,0.0001267009,0.000206575,0.0001674169],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8523803,"threshold_uncertainty_score":0.9996582,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02806232812616088,"score_gpt":0.2553470633048817,"score_spread":0.2272847351787208,"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."}}