{"id":"W4411950667","doi":"10.1109/forge66646.2025.00011","title":"Automated Codebase Reconciliation using Large Language Models","year":2025,"lang":"en","type":"article","venue":"","topic":"Machine Learning in Bioinformatics","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Advanced Micro Devices (Canada); University of Toronto","funders":"","keywords":"Codebase; Computer science; Programming language; Software","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":[0.006520343,0.00142571,0.001022983,0.002964331,0.001150978,0.003531401,0.003658595,0.001493454,0.002153609],"category_scores_gemma":[0.03212935,0.001149191,0.002025441,0.00161666,0.0008967372,0.003655009,0.003351044,0.002594104,0.001921999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001831745,"about_ca_system_score_gemma":0.004422639,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00533453,"about_ca_topic_score_gemma":0.01217947,"domain_scores_codex":[0.99444,0.002120506,0.0004260782,0.001076194,0.001731535,0.0002058498],"domain_scores_gemma":[0.9831411,0.007946114,0.00173608,0.003944254,0.002908216,0.0003242785],"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":[0.0004651648,0.0006572392,0.01441365,0.0007513413,0.0003229603,0.000860349,0.002433736,0.2537925,0.03912584,0.01844509,0.01830273,0.6504294],"study_design_scores_gemma":[0.00003510995,0.00007429127,0.0009021777,0.00006378852,0.00005348836,0.0001483827,0.0002082603,0.960738,0.01539206,0.0131455,0.009187305,0.00005170114],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02368472,0.0001941212,0.9386151,0.0005828668,0.00005970252,0.0002637824,0.0005589473,0.03507191,0.0009688176],"genre_scores_gemma":[0.1487321,0.0001366551,0.8438538,0.0002691595,0.0000298468,0.0003204644,0.002774691,0.00246388,0.001419411],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006520343,"threshold_uncertainty_score":0.03448331,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009332571674709807,"score_gpt":0.3005197324523237,"score_spread":0.2911871607776139,"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."}}