{"id":"W3179755228","doi":"10.1109/sbst52555.2021.00019","title":"SWAT tool at the SBST 2021 Tool Competition","year":2021,"lang":"en","type":"article","venue":"","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Context (archaeology); Competition (biology); Test (biology); Computer science; Reliability engineering; Simulation; 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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00007046941,0.00008088692,0.00008661699,0.00001371186,0.0001175837,0.00001376732,0.0000905369,0.0001043581,0.005585011],"category_scores_gemma":[0.00001319479,0.0000612564,0.00004597016,0.0001082837,0.00004511034,0.00005189922,0.00007157968,0.0001574045,0.001142682],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006330938,"about_ca_system_score_gemma":0.00001250819,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003147353,"about_ca_topic_score_gemma":0.0001324739,"domain_scores_codex":[0.9995366,0.00001383114,0.0001242908,0.000110684,0.00006012067,0.0001545228],"domain_scores_gemma":[0.999613,0.00004469881,0.000008560103,0.000300628,0.00002065498,0.00001245189],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000287704,0.0001354265,0.006598857,0.000135203,0.0003766456,0.0002419853,0.0005694506,0.0283106,0.1420886,0.5521724,0.08576176,0.1835804],"study_design_scores_gemma":[0.000642097,0.00002591793,0.02353983,0.00002826518,0.0000356841,0.0001377411,0.0003802567,0.03313611,0.3177899,0.003952243,0.6197793,0.0005526154],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8490195,0.0007003206,0.01153893,0.00351101,0.000430495,0.000131395,0.00001280786,0.0008141532,0.1338413],"genre_scores_gemma":[0.9900771,0.0001499384,0.0004320593,0.0002812756,0.00005370728,0.00001386986,0.00002678769,0.00001317479,0.008952075],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5482202,"threshold_uncertainty_score":0.999635,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.004745931818835941,"score_gpt":0.1789288905711974,"score_spread":0.1741829587523615,"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."}}