Reflective Prompt Optimization of a Deep-Research Agent: A Topic-Disjoint, Multi-Seed Protocol for Measuring Transferable Gains and Prompt-Level Overfitting
Computer Science > Computation and Language arXiv:2610.12601 (cs) [Submitted on 8 Oct 2026] Title: Reflective Prompt Optimization of a Deep-Research Agent: A Topic-Disjoint, Multi-Seed Protocol for Measuring Transferable Gains and Prompt-Level Overfitting Authors: Elias Lumer, Sahil Sen, Vamse Kumar Subbiah, Kevin Paul.
计算机科学 > 计算与语言 arXiv:2610.12601 (cs) [2026年10月8日提交] 标题：深度研究智能体的反射式提示词优化：一种用于衡量可迁移增益和提示词级过拟合的主题分离、多种子协议 作者：Elias Lumer, Sahil Sen, Vamse Kumar Subbiah, Kevin Paul。
Abstract: Prompt optimization is an attractive alternative to reinforcement learning for adapting language-model agents, since it is cheap and produces human-readable artifacts. On factual tasks, however, it carries a failure mode that scalar-reward reinforcement learning does not. An optimizer that reads the training examples can write their answers into the prompt, so an in-distribution gain may be memorization rather than transferable behaviour.
摘要：提示词优化是强化学习的一种极具吸引力的替代方案，用于适配语言模型智能体，因为它成本低廉且能产生人类可读的产物。然而，在事实性任务中，它存在一种标量奖励强化学习所没有的失效模式。读取训练示例的优化器可能会将答案写入提示词中，因此分布内的增益可能是记忆而非可迁移的行为。
We apply reflective prompt evolution (GEPA) to the two system prompts of a multi-agent deep-research agent and evaluate it under a topic-disjoint split in which no topic is shared across train, validation, and test. Held-out citation reward rises by 0.079, from 0.702 to 0.781, with a bootstrap 95% confidence interval of [0.051, 0.107], and the gain replicates across three seeds (mean 0.107).
我们将反射式提示词演化（GEPA）应用于多智能体深度研究系统的两个系统提示词，并在主题分离的划分下对其进行评估，确保训练集、验证集和测试集之间没有共享主题。留出引用奖励提高了0.079，从0.702上升至0.781，自助法95%置信区间为[0.051, 0.107]，且该增益在三个种子中均可复现（平均值为0.107）。
Nearly half of the winning prompt is a memorized answer key, yet a controlled ablation shows this content earns no reward even on the exact training questions it encodes. The large validation-minus-held-out gap is therefore small-sample selection under a noisy judge, not memorization. Varying the verifier isolates what makes the gain transfer.
获胜提示词中近一半是记忆的答案键，但受控消融实验表明，即使在它所编码的确切训练问题上，这些内容也无法获得奖励。因此，验证集与留出集之间巨大的差距是噪声评估器下的样本选择偏差，而非记忆。改变验证器可以分离出促成增益迁移的因素。
Decomposed per-dimension feedback is the causal ingredient, and collapsing it to a single scalar reward makes held-out transfer negative. The gain persists under an out-of-family judge and is largely reproduced by a hand-written version of the discovered recipe. Credible evaluation of prompt optimizers therefore requires held-out topics, multiple seeds, and verifier ablations.
分解后的逐维度反馈是因果要素，将其压缩为单一标量奖励会导致留出迁移变为负值。这种增益在非同源评估器下依然存在，并且可以通过手写版本的发现方案在很大程度上复现。因此，对提示词优化器的可信评估需要留出主题、多个种子以及验证器消融实验。