{"id":"W1980089213","doi":"10.1109/compsac.2012.23","title":"Evaluating Reliability-Testing Usage Models","year":2012,"lang":"en","type":"article","venue":"","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Reliability (semiconductor); Computer science; Reliability engineering; Goodness of fit; Markov chain; Markov model; Non-regression testing; Software quality; Markov process; Data mining; Machine learning; Statistics; Software; Engineering; Mathematics","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":[0.004452098,0.0001113273,0.0001595283,0.00009508201,0.0002115912,0.0001599521,0.0007991947,0.00005483513,0.00007580212],"category_scores_gemma":[0.00224533,0.00008956841,0.00009111459,0.0009712772,0.00004884851,0.001692898,0.000514769,0.0001970298,0.0002032984],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007253727,"about_ca_system_score_gemma":0.00008350309,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001653412,"about_ca_topic_score_gemma":0.000001148704,"domain_scores_codex":[0.9977809,0.0002753207,0.0002798967,0.0003685691,0.000699986,0.0005952823],"domain_scores_gemma":[0.9972606,0.001325012,0.00004854205,0.0008773556,0.0002583484,0.000230118],"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":[0.000006527797,0.0006781311,0.09663051,0.0001140336,0.00005009624,0.000003740334,0.002283297,0.09269369,0.003471667,0.1235683,0.0006382578,0.6798618],"study_design_scores_gemma":[0.00008376753,0.00003697014,0.001522092,0.000008981729,0.000004494515,0.000004496294,0.00001983656,0.9770011,0.0006160593,0.02038223,0.0001911076,0.0001288967],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1515516,0.0001341882,0.8334702,0.0004397359,0.0000951341,0.0001205216,2.75777e-7,0.0003584719,0.01382986],"genre_scores_gemma":[0.6650276,0.000002532286,0.3344401,0.00009913299,0.00006664017,0.00001267543,4.399025e-7,0.00000446527,0.0003464651],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8843074,"threshold_uncertainty_score":0.3652494,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2197938598611162,"score_gpt":0.4020445771824858,"score_spread":0.1822507173213697,"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."}}