{"id":"W1984953175","doi":"10.1109/tse.2014.2361131","title":"Replicating and Re-Evaluating the Theory of Relative Defect-Proneness","year":2014,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Polytechnique Montréal; Queen's University","funders":"","keywords":"Computer science; Context (archaeology); Software quality; Replication (statistics); Software; Source code; Software bug; Code review; Quality (philosophy); Software system; Data science; Software development; Software engineering; Statistics; 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":["metaresearch"],"category_scores_codex":[0.2518065,0.003079766,0.00311778,0.008656621,0.00184258,0.007104244,0.01180619,0.004180531,0.004762676],"category_scores_gemma":[0.6205637,0.0009179334,0.004505389,0.005433595,0.01577817,0.01591381,0.005980115,0.009646421,0.001384562],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006406179,"about_ca_system_score_gemma":0.005826174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009191635,"about_ca_topic_score_gemma":0.003685676,"domain_scores_codex":[0.8297715,0.117249,0.00621735,0.01807806,0.02740077,0.001283344],"domain_scores_gemma":[0.250043,0.5635868,0.0189993,0.1269177,0.03836402,0.002089092],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.001290215,0.0007503312,0.0741566,0.004240292,0.006902615,0.0007914178,0.01481295,0.04776968,0.002672903,0.4413624,0.01473963,0.390511],"study_design_scores_gemma":[0.0006068012,0.002393692,0.03082747,0.003091214,0.001958291,0.000585589,0.004872724,0.07900339,0.003922858,0.8239341,0.0482787,0.0005251677],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2104832,0.03232577,0.6631627,0.04281745,0.01093812,0.002024701,0.001241873,0.001385922,0.03562027],"genre_scores_gemma":[0.8173339,0.007015762,0.1577797,0.008819358,0.003021281,0.001614762,0.0006239606,0.0007286209,0.003062498],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7481935,"threshold_uncertainty_score":0.9226558,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02886620484135756,"score_gpt":0.2784184521015144,"score_spread":0.2495522472601568,"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."}}