{"id":"W3009853975","doi":"10.1002/smr.2250","title":"Guidelines for evaluating bug‐assignment research","year":2020,"lang":"en","type":"article","venue":"Journal of Software Evolution and Process","topic":"Software Engineering Research","field":"Computer Science","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Faculty of Graduate Studies and Research, University of Alberta; Alberta Innovates - Technology Futures","keywords":"Computer science; Metric (unit); Ranking (information retrieval); Empirical research; Task (project management); Set (abstract data type); Data science; Data mining; Information retrieval; Statistics; Systems engineering; Mathematics; Engineering; Operations management","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003155834,0.00009604596,0.0001827564,0.0001885144,0.0001899232,0.0001489551,0.0006124459,0.00005875265,0.000006824338],"category_scores_gemma":[0.02109937,0.00007970992,0.00005871441,0.0005886291,0.00004529506,0.0005050522,0.0001382349,0.0003298723,0.000004220552],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001015626,"about_ca_system_score_gemma":0.0004523297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002052055,"about_ca_topic_score_gemma":2.11195e-7,"domain_scores_codex":[0.9978871,0.00008522062,0.0004500673,0.0002183226,0.001064112,0.0002952187],"domain_scores_gemma":[0.9943646,0.001150784,0.0001592519,0.0001280987,0.003934638,0.0002626363],"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":[0.001055755,0.0004316932,0.1044556,0.004634228,0.000413883,0.0001164642,0.01713027,0.05593568,0.008346365,0.008832038,0.2306016,0.5680464],"study_design_scores_gemma":[0.009521442,0.01218475,0.04397045,0.001964713,0.00008907748,0.0007459687,0.002382383,0.7953093,0.008018774,0.0755024,0.04874389,0.001566867],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02168723,0.002358507,0.9672332,0.008207779,0.0002393693,0.0001948881,0.000002016308,0.00007150811,0.000005467642],"genre_scores_gemma":[0.7489992,0.00005907121,0.2497695,0.0003551865,0.0007183036,0.00002335612,5.840165e-7,0.00001664063,0.00005822544],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7393736,"threshold_uncertainty_score":0.9871463,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2884186218319023,"score_gpt":0.4726799934303204,"score_spread":0.1842613715984182,"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."}}