{"id":"W4402571307","doi":"10.1109/icstw60967.2024.00021","title":"Annotating Control-Flow Graphs for Formalized Test Coverage Criteria","year":2024,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; Queen's University","funders":"","keywords":"Computer science; Control flow; Test (biology); Control flow graph; Control (management); Programming language; Software engineering; Artificial intelligence","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":[0.000511543,0.0001391553,0.000159469,0.0001286791,0.0001367733,0.0006529616,0.000425019,0.00004971929,0.00001525661],"category_scores_gemma":[0.000745773,0.0001144756,0.0001083591,0.0003094828,0.00002296146,0.0005178344,0.00007937395,0.00008212336,0.00001531415],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001902505,"about_ca_system_score_gemma":0.0000491619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002447901,"about_ca_topic_score_gemma":0.000001141841,"domain_scores_codex":[0.9989805,0.00002297649,0.0002219841,0.0003419202,0.0001372059,0.0002954373],"domain_scores_gemma":[0.9974195,0.00206813,0.0000324912,0.0003290871,0.00008875485,0.00006197459],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001985659,0.00009336557,0.002088393,0.0003142184,0.00005304465,0.0000744827,0.0007964584,0.00004128351,0.002241082,0.1252569,0.5503671,0.3186539],"study_design_scores_gemma":[0.000339431,0.0001690245,0.0001183003,0.00007932176,0.000007573854,0.00003370527,0.000001060086,0.7553169,0.0006570732,0.2367471,0.006355046,0.0001754682],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0004183085,0.0002031483,0.9746491,0.0009587075,0.0004153927,0.0002532175,0.00002177159,0.02150341,0.001576901],"genre_scores_gemma":[0.4508608,0.000003517524,0.5477678,0.001019029,0.0000464556,0.00008480016,0.00000415025,0.00001276544,0.0002006279],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7552757,"threshold_uncertainty_score":0.6296524,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01748680288172474,"score_gpt":0.2954803116486411,"score_spread":0.2779935087669164,"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."}}