{"id":"W2162836425","doi":"10.1109/issre.1991.145373","title":"Metrics evaluation of software reliability growth models","year":2002,"lang":"en","type":"article","venue":"","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Metric (unit); Reliability (semiconductor); Computer science; Software quality; Software metric; Field (mathematics); Software; Reliability engineering; Predictive validity; Term (time); Data mining; Machine learning; Software development; Statistics; Mathematics; Engineering; Programming language","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":[0.01498851,0.0013844,0.001157479,0.005799826,0.0004380884,0.002548089,0.001384577,0.0009475536,0.0008511896],"category_scores_gemma":[0.0977207,0.0004281817,0.001034849,0.003576941,0.000631091,0.002907693,0.001413618,0.0008290242,0.000235252],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002842211,"about_ca_system_score_gemma":0.001826545,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00487424,"about_ca_topic_score_gemma":0.003863835,"domain_scores_codex":[0.9826231,0.009930815,0.001185026,0.0007072291,0.005173244,0.0003804729],"domain_scores_gemma":[0.9451629,0.03248115,0.004484976,0.003999915,0.01325914,0.0006118772],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002065534,0.0001649056,0.02404973,0.0003606237,0.0002651961,0.0001556351,0.0004182652,0.6760697,0.002450544,0.02852765,0.00294515,0.2643861],"study_design_scores_gemma":[0.000009412486,0.0001515777,0.002220793,0.00005107185,0.00002906558,0.00005296501,0.00004948355,0.9873751,0.001426493,0.007151692,0.00145734,0.00002487209],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2212471,0.003935251,0.7608863,0.001161269,0.0001550043,0.0003664157,0.0007299245,0.002755246,0.008763544],"genre_scores_gemma":[0.8212199,0.000972487,0.1747209,0.00006550476,0.00005889514,0.0003439165,0.001406508,0.0002455743,0.0009662829],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01498851,"threshold_uncertainty_score":0.0792678,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1057111148427386,"score_gpt":0.3088034808344209,"score_spread":0.2030923659916823,"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."}}