ThinkReset: Learnable Intermediate Interface Construction for Bounded-Context Long-Horizon Reasoning

ThinkReset：面向有限上下文长程推理的可学习中间接口构建

Abstract: Long chain-of-thought reasoning improves performance on complex problems, but it also introduces redundancy accumulation, context overflow, and error anchoring. 摘要： 长思维链推理提升了模型处理复杂问题的能力，但也带来了冗余积累、上下文溢出以及错误锚定等问题。

We argue that under bounded context windows, the core bottleneck is not trajectory compression or test-time control, but the absence of a reusable intermediate interface that can replace discarded history and support continued solving. 我们认为，在有限的上下文窗口下，核心瓶颈并非轨迹压缩或测试时控制，而是缺乏一种可重用的中间接口，以替代被丢弃的历史信息并支持持续求解。

We further identify a key failure mode of outcome-reward-driven long-chain reinforcement learning: when the model has not solved the task before the window is nearly exhausted, the final-answer reward encourages premature guessing rather than continued careful reasoning. 我们进一步识别出一种由结果奖励驱动的长链强化学习的关键失效模式：当模型在上下文窗口即将耗尽前仍未解决任务时，最终答案的奖励会诱导模型进行过早的猜测，而非继续进行严谨的推理。

We propose ThinkReset, a text-space instantiation of this view. ThinkReset explicitly constructs reusable intermediate interfaces through interface writeback and reset, and directly optimizes post-reset continuation success. 我们提出了 ThinkReset，这是上述观点在文本空间的一种实例化实现。ThinkReset 通过接口回写（writeback）和重置（reset）显式地构建可重用的中间接口，并直接优化重置后的持续求解成功率。

Across multiple long-horizon reasoning benchmarks, this perspective consistently improves success rates under fixed context windows. 在多个长程推理基准测试中，该方法在固定上下文窗口下持续提升了任务成功率。