{"id":"W4402527462","doi":"10.1145/3678720.3685320","title":"Abstract Debugging with GobPie","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","funders":"","keywords":"Debugging; Computer science; Programming language; Software engineering","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":[0.0001630542,0.00008401339,0.00006135493,0.00008894316,0.00005166715,0.0003738855,0.0003520185,0.00002509289,0.00002461369],"category_scores_gemma":[0.00002766423,0.00005599681,0.0000238915,0.0003032144,0.00001890095,0.0003204474,0.00007406858,0.0001042069,0.0001208713],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001907211,"about_ca_system_score_gemma":0.00006513948,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006012093,"about_ca_topic_score_gemma":0.000003631098,"domain_scores_codex":[0.9993421,0.000005171356,0.0000799029,0.0002608526,0.0001447538,0.0001671607],"domain_scores_gemma":[0.9993942,0.0002064652,0.00001060094,0.0003183139,0.00002308961,0.00004735589],"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":[0.000002327888,0.00003391004,0.001957061,0.00008725515,0.0000365114,0.0005065418,0.0004263297,0.00003439016,0.0002291242,0.1194409,0.1902744,0.6869713],"study_design_scores_gemma":[0.0003063463,0.0004380193,0.01968326,0.001659437,0.0000301833,0.001081106,0.00001049651,0.4455892,0.0105193,0.4449759,0.07416385,0.001542838],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.00157855,0.0003077346,0.9093209,0.001003888,0.0001326412,0.0000457134,2.146237e-7,0.05920551,0.02840489],"genre_scores_gemma":[0.6811732,0.000001969865,0.3182503,0.0002122334,0.00003031988,0.000006489856,3.021652e-7,0.000007317293,0.0003178346],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6854285,"threshold_uncertainty_score":0.3605387,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01472623548697537,"score_gpt":0.2515278555332744,"score_spread":0.236801620046299,"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."}}