{"id":"W2867457158","doi":"10.1145/3213846.3213847","title":"Safe and sound program analysis with Flix","year":2018,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Sound (geography); Acoustics; Physics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.0001398758,0.00006819502,0.00009305578,0.0001135196,0.00009356119,0.0001338096,0.0002337049,0.00002223341,0.00001013128],"category_scores_gemma":[0.00003329973,0.00004550674,0.00001951226,0.0007716225,0.0001006785,0.0001276385,0.0001046229,0.00003436092,0.00001037137],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000006069999,"about_ca_system_score_gemma":0.00001174696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000104053,"about_ca_topic_score_gemma":0.00006652671,"domain_scores_codex":[0.9994263,0.00001373565,0.00006984327,0.00024045,0.0001135402,0.0001361291],"domain_scores_gemma":[0.9994016,0.00006980111,0.00002771265,0.0003682142,0.000078896,0.00005374496],"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":[0.000009364639,0.0001321446,0.4631858,0.000009959478,0.0002672798,0.00001675696,0.0007512298,0.000003505627,0.0000102942,0.01798155,0.01030726,0.5073248],"study_design_scores_gemma":[0.0006416073,0.004187991,0.2172571,0.00004711693,0.0003839611,0.0001744498,0.00001781449,0.5381323,0.00167832,0.2239272,0.01238035,0.0011719],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03251253,0.00001690185,0.950149,0.0001816915,0.00001586616,0.00007543447,1.344628e-7,0.009879443,0.007169017],"genre_scores_gemma":[0.4974201,3.873481e-7,0.5023146,0.0001260446,0.00001548475,0.000007151295,2.582776e-7,0.000001833784,0.0001141783],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5381287,"threshold_uncertainty_score":0.1855711,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0160174123335903,"score_gpt":0.2840345725565023,"score_spread":0.2680171602229121,"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."}}