{"id":"W6929267437","doi":"10.48550/arxiv.2507.06762","title":"Detecção de Conflitos Semânticos com Testes Gerados por LLM","year":2025,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Amino Acid Enzymes and Metabolism","field":"Biochemistry, Genetics and Molecular Biology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Merge (version control); Unit testing; Codebase; Benchmark (surveying); Code (set theory); Test case","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":[0.00802588,0.001859129,0.001058545,0.003246417,0.0009093531,0.002778581,0.002737991,0.002355632,0.004299751],"category_scores_gemma":[0.06260961,0.001180601,0.001765772,0.001427324,0.002044123,0.004822586,0.003760153,0.00214798,0.001177077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001399818,"about_ca_system_score_gemma":0.002264596,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002757798,"about_ca_topic_score_gemma":0.004647836,"domain_scores_codex":[0.9905238,0.002755536,0.0006531258,0.001571474,0.003781852,0.0007142796],"domain_scores_gemma":[0.9384453,0.04163522,0.004739848,0.009554188,0.004643167,0.000982206],"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":[0.00376264,0.001114694,0.1784307,0.002228557,0.0005270482,0.007741131,0.004292813,0.1001307,0.09263543,0.04341354,0.0235164,0.5422064],"study_design_scores_gemma":[0.0002459028,0.0005458755,0.01072893,0.000259155,0.0002355783,0.001927646,0.0008763146,0.7181104,0.1990483,0.03994375,0.02791443,0.0001636863],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3187738,0.000729534,0.568719,0.001179906,0.0002722082,0.0007634698,0.002315122,0.09864767,0.008599392],"genre_scores_gemma":[0.6823803,0.0001382573,0.3078731,0.0004249553,0.00004985681,0.0003737158,0.002001581,0.004506315,0.002252019],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.00802588,"threshold_uncertainty_score":0.04244542,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0346961874350509,"score_gpt":0.1941688125350722,"score_spread":0.1594726251000213,"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."}}