{"id":"W3086099593","doi":"10.14778/3407790.3407856","title":"Sentinel","year":2020,"lang":"en","type":"article","venue":"Proceedings of the VLDB Endowment","topic":"Software System Performance and Reliability","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Debugging; Computer science; Variety (cybernetics); Server; Distributed computing; Web application; Operating system; Database; Embedded system; Artificial intelligence","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":[0.0001719978,0.00008483796,0.0001250777,0.00001629563,0.00006792759,0.00004200329,0.001011274,0.00002653765,0.00000617559],"category_scores_gemma":[0.00007001482,0.0000500982,0.00009418442,0.0003329955,0.00003520961,0.0002380422,0.0005036266,0.00008104108,0.000028456],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001993528,"about_ca_system_score_gemma":0.00002028679,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006726887,"about_ca_topic_score_gemma":4.948282e-8,"domain_scores_codex":[0.9991226,0.000003204711,0.0002082948,0.0002149659,0.0002981702,0.0001527371],"domain_scores_gemma":[0.9995416,0.00001280229,0.0001340668,0.0001349166,0.0001131456,0.00006351549],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00006989229,0.0004437634,0.569014,0.002120252,0.0002001888,0.000001756519,0.01752236,0.0001156577,0.1784233,0.1266838,0.08076434,0.02464074],"study_design_scores_gemma":[0.001899311,0.0002846804,0.07463063,0.0002648226,0.00004835024,0.00003454602,0.0005030803,0.02723593,0.8416666,0.007930712,0.04484654,0.0006548392],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.945706,0.0001802091,0.003305895,0.03199699,0.0006601915,0.000688605,0.000001590462,0.0003752715,0.01708525],"genre_scores_gemma":[0.9970093,0.000008723503,0.002009199,0.0007967401,0.00007775448,0.00001446495,5.981876e-8,0.00000401066,0.00007978886],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6632432,"threshold_uncertainty_score":0.2042945,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01366508124302437,"score_gpt":0.2022727780473852,"score_spread":0.1886076968043608,"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."}}