{"id":"W6912068948","doi":"10.5281/zenodo.15605218","title":"Performance Dataset for Hardware Model Checking on Btor2 Benchmarks (Technical Report, May 2025)","year":2025,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Benchmark (surveying); Set (abstract data type); Construct (python library); Scripting language; Selection (genetic algorithm)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00255982,0.003310438,0.001012273,0.00203898,0.0007862846,0.001340668,0.003573287,0.001696734,0.01023855],"category_scores_gemma":[0.008978052,0.0005978884,0.001484671,0.003329387,0.000771863,0.001718591,0.001404355,0.002134653,0.008120517],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001839769,"about_ca_system_score_gemma":0.002514168,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01494372,"about_ca_topic_score_gemma":0.02208032,"domain_scores_codex":[0.9958728,0.0007741531,0.0003804266,0.0007651492,0.001733605,0.0004738725],"domain_scores_gemma":[0.9942971,0.001534358,0.000391655,0.001695847,0.001798354,0.0002828126],"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":[0.002068716,0.0007330661,0.007780056,0.002116898,0.0003468684,0.0002522785,0.00008429054,0.07135491,0.00881334,0.004345307,0.8573789,0.04472532],"study_design_scores_gemma":[0.002376878,0.002023406,0.04029656,0.0005232808,0.0003402385,0.0007871446,0.0002425555,0.4497636,0.05002696,0.02167542,0.4316469,0.0002971441],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"empirical","genre_scores_codex":[0.1204804,0.003317776,0.02644444,0.001497725,0.0008420408,0.0007007251,0.7313823,0.08730485,0.02802969],"genre_scores_gemma":[0.08996532,0.0003637139,0.01874037,0.0003977946,0.00006363961,0.0005128657,0.8833262,0.003013591,0.003616575],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01494372,"threshold_uncertainty_score":0.03425139,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03964419333641551,"score_gpt":0.2992744602368088,"score_spread":0.2596302669003933,"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."}}