{"id":"W4416408255","doi":"10.48550/arxiv.2507.19219","title":"How Much Do Large Language Model Cheat on Evaluation? Benchmarking Overestimation under the One-Time-Pad-Based Framework","year":2025,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Instituto de Ciencias del Mar y Limnología, Universidad Nacional Autónoma de México; Innovation and Technology Fund; Institute for Catastrophic Loss Reduction; University of Science and Technology of China; Microsoft Research; National Natural Science Foundation of China; Hong Kong Polytechnic University","keywords":"Benchmarking; Benchmark (surveying); Key (lock); Quality (philosophy); Test (biology); Measure (data warehouse)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04034681,0.001893155,0.001658763,0.002051446,0.001072281,0.004985921,0.004147925,0.002337679,0.003990305],"category_scores_gemma":[0.1911555,0.0008003547,0.001246108,0.002010519,0.003374164,0.009788502,0.005706657,0.004138034,0.003214247],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002986988,"about_ca_system_score_gemma":0.004753126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002641275,"about_ca_topic_score_gemma":0.003808315,"domain_scores_codex":[0.9296769,0.04316965,0.00447539,0.005212394,0.01497769,0.002487893],"domain_scores_gemma":[0.871769,0.06280755,0.005099596,0.04622443,0.01127946,0.002819849],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003966946,0.001261786,0.0510418,0.002295468,0.001049616,0.0007150833,0.001609017,0.1784524,0.02989437,0.07987857,0.1096194,0.5402154],"study_design_scores_gemma":[0.0007776256,0.002600469,0.01003155,0.0006599252,0.0002443146,0.0009515359,0.0008783797,0.769046,0.05651334,0.1187792,0.03925912,0.0002585007],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4195424,0.01075645,0.4569865,0.01463692,0.002500035,0.0009993317,0.004449752,0.05353646,0.03659216],"genre_scores_gemma":[0.8490027,0.0007073244,0.1359824,0.002359886,0.0002864154,0.0005119403,0.004140669,0.003980286,0.003028422],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9596532,"threshold_uncertainty_score":0.2133768,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08765754670620739,"score_gpt":0.2418694500749871,"score_spread":0.1542119033687797,"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."}}