{"id":"W2200054515","doi":"10.1007/s10664-015-9409-1","title":"On the unreliability of bug severity data","year":2015,"lang":"en","type":"article","venue":"Empirical Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":51,"is_retracted":false,"has_abstract":false,"ca_institutions":"Queen's University; University of Waterloo","funders":"","keywords":"Computer science; Software bug; Reliability (semiconductor); Eclipse; Software; Data mining; Data science; 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09695513,0.0008170771,0.001630325,0.009304969,0.001686389,0.003494539,0.003583395,0.003928188,0.00297214],"category_scores_gemma":[0.6533794,0.001445146,0.0013667,0.008741851,0.005241038,0.007333129,0.003139205,0.005455018,0.0006543326],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002349053,"about_ca_system_score_gemma":0.00192291,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006121226,"about_ca_topic_score_gemma":0.004571605,"domain_scores_codex":[0.8796456,0.07537611,0.009241302,0.01487657,0.01908913,0.001771282],"domain_scores_gemma":[0.1072407,0.8046948,0.0278662,0.0482057,0.01117944,0.0008131831],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001035485,0.00030238,0.6310661,0.001294127,0.002069394,0.001733325,0.00408586,0.105813,0.001272804,0.08234027,0.01992911,0.1490582],"study_design_scores_gemma":[0.0003688539,0.0003717462,0.2513094,0.002063309,0.0009832568,0.003846496,0.002148798,0.3349619,0.004609265,0.3798098,0.01925012,0.0002771238],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6203692,0.005044597,0.3304657,0.02043796,0.0007031248,0.0003131246,0.006355788,0.001872108,0.01443835],"genre_scores_gemma":[0.9758321,0.0004980472,0.01894347,0.0012159,0.000351724,0.00009840465,0.002261396,0.0002111839,0.0005877369],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9030449,"threshold_uncertainty_score":0.5127538,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1072583466602886,"score_gpt":0.3248063854558618,"score_spread":0.2175480387955732,"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."}}