{"id":"W2890687291","doi":"","title":"[Journal First] Are Fix-Inducing Changes a Moving Target?: A Longitudinal Case Study of Just-in-Time Defect Prediction","year":2018,"lang":"en","type":"article","venue":"International Conference on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Code (set theory); Brier score; Interpretation (philosophy); Calibration; Artificial intelligence; Statistics; Mathematics; 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":[0.007066893,0.0003600919,0.0004239078,0.001603532,0.001079066,0.001468782,0.001233634,0.001337326,0.001650271],"category_scores_gemma":[0.05089667,0.0003365959,0.0005789544,0.001417606,0.001058861,0.002704854,0.001179235,0.00156036,0.0007315044],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001301351,"about_ca_system_score_gemma":0.0009665978,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01442295,"about_ca_topic_score_gemma":0.01719102,"domain_scores_codex":[0.9968995,0.001131492,0.0002381541,0.0006845146,0.0008074596,0.0002388704],"domain_scores_gemma":[0.9549237,0.02036821,0.007125095,0.007740305,0.007897886,0.001944716],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000364771,0.0007151933,0.8796367,0.00009961771,0.0001477411,0.00149276,0.003386198,0.0199508,0.001389299,0.002148672,0.007861974,0.0828063],"study_design_scores_gemma":[0.0001043547,0.002417191,0.6154879,0.000272073,0.0001868356,0.003520816,0.004865294,0.3335606,0.007032293,0.01322576,0.01908208,0.0002447888],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9870603,0.0003025887,0.008566041,0.001287113,0.00005300501,0.00007801202,0.0008532564,0.0002027047,0.001596944],"genre_scores_gemma":[0.9933501,0.0001288447,0.004124141,0.0001697033,0.00003296202,0.00003967699,0.00125007,0.00005995442,0.0008444988],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01442295,"threshold_uncertainty_score":0.03737372,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05035473449789288,"score_gpt":0.2965110632208014,"score_spread":0.2461563287229085,"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."}}