{"id":"W4394746056","doi":"10.1145/3597503.3639112","title":"Mozi: Discovering DBMS Bugs via Configuration-Based Equivalent Transformation","year":2024,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"China Postdoctoral Science Foundation; National Natural Science Foundation of China","keywords":"Computer science; Correctness; Programming language; SQL; Semantics (computer science); Database; Query language; Transformation (genetics); Task (project management)","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002552114,0.0001016661,0.00008025038,0.0001097246,0.00009447744,0.0002812263,0.0002907765,0.00004259543,0.00003266153],"category_scores_gemma":[0.00002945519,0.00008352132,0.00005365042,0.0002701835,0.00001838402,0.0006305156,0.00002058312,0.0001050151,0.00008328218],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005432488,"about_ca_system_score_gemma":0.00006790756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001062846,"about_ca_topic_score_gemma":0.000004544919,"domain_scores_codex":[0.9991687,0.00002817292,0.0001972284,0.0002243641,0.0002091874,0.00017233],"domain_scores_gemma":[0.999351,0.0002767244,0.00002178956,0.0002669189,0.00004043256,0.00004313392],"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":[0.00001215286,0.0001049065,0.0006435403,0.0005516041,0.0000514548,0.00006010525,0.008605608,0.004338082,0.01009784,0.3016053,0.02856987,0.6453596],"study_design_scores_gemma":[0.000076786,0.00005020738,0.0001551392,0.0001302253,0.000004334761,0.000006634438,0.000004904077,0.9608776,0.01487693,0.02146072,0.002204089,0.0001524547],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002005401,0.00007330566,0.9739308,0.001524924,0.0003158663,0.0001387346,8.978207e-7,0.01611729,0.00589273],"genre_scores_gemma":[0.9274188,0.000003245155,0.07194069,0.0003606151,0.00004227678,0.00002596075,0.000005840139,0.000007902424,0.0001946334],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9565395,"threshold_uncertainty_score":0.3405901,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01918079054616061,"score_gpt":0.2707753826562442,"score_spread":0.2515945921100836,"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."}}