不更新参数，让AI解题更准更稳。
CP-Agent: A Calibrated Risk-Controlled Agent for Feedback-Driven Competitive Programming
- 用反馈构建校准机制，控制错误提交风险。
- 在真实竞赛数据集上，通过率从25.8%提至48.5%。
- 适合想提升AI编程可靠性又不想重训的开发者。
大型语言模型在竞赛级编程任务中仍表现不足，现有代理方法多依赖大量推理采样或昂贵的多阶段后训练。本文研究执行反馈何时有效，并揭示其增益机制。将反馈驱动求解建模为校准停止过程，识别出三个关键量：误接入风险、否定劣质程序的证据强度、活跃状态的成功危险性。在保留轨迹校准和预声明有限控制器集合的前提下，所生成的结构化证书可下界估计无误接入情况下的纯净成功率。据此设计双粒度验证、测试增强与经验驱动自演化三类机制，构建出无需参数更新的CP-Agent。在LiveCodeBench Pro上，Pass@1从25.8%提升至48.5%；在ICPC-Eval上，Refine@5提升11.0%。三种主流大模型骨架均验证其处于成本-精度效率前沿，消融实验表明各组件分别主导对应证书量的变化。
原文摘要 · Abstract (English)
Large language models still struggle with contest-level programming, while many agentic remedies rely on massive inference-time sampling or expensive multi-stage post-training. We study when execution feedback reliably helps an LLM CP solver and which mechanisms govern the gains. We model feedback-driven solving as a calibrated stopped process and identify three quantities: false-admission risk, program-level evidence against bad programs, and the active-state success hazard. Under held-out trace calibration and selection from a pre-declared finite controller manifest, the resulting structural certificate lower-bounds the clean success probability before false admission. We instantiate mechanisms targeting these quantities as Dual-Granularity Verification, Test Augmentation, and Experience-Driven Self-Evolving, yielding CP-Agent. Without updating any parameters, CP-Agent raises Pass@1 from 25.8\% to 48.5\% on LiveCodeBench Pro and improves Refine@5 by 11.0\% on ICPC-Eval. Across three LLM backbones, CP-Agent lies on the cost--accuracy efficiency frontier, and ablations show that each component primarily affects its corresponding certificate quantity.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。