arXiv：cs.LG· Sri Vatsa Vuddanti, Satwik Kumar Chittiprolu·· 5 小时前AI 评分40
工具增强型智能体的可恢复性定律：ERR 度量
Recoverability Has a Law: The ERR Measure for Tool-Augmented Agents
AI 导读
研究者提出 Expected Recovery Regret（ERR）度量，量化恢复策略在随机执行噪声下偏离最优策略的程度，并推导出 ERR 与经验可观测量 Efficiency Score（ES）之间的一阶定量定律。
正文
Abstract:Language model agents often appear capable of self-recovery after failing tool call executions, yet this behavior lacks a formal explanation. We present a predictive theory that resolves this gap by showing that recoverability follows a measurable law. To elaborate, we formalize recoverability through Expected Recovery Regret (ERR), which quantifies the deviation of a recovery policy from the optimal one under stochastic execution noise, and derive a first-order relationship between ERR and an empirical observable quantity, the Efficiency Score (ES). This yields a falsifiable first-order quantitative law of recovery dynamics in tool-using agents. We empirically validate the law across five tool-use benchmarks spanning controlled perturbations, diagnostic reasoning, and real-world APIs. Across model scales, perturbation regimes, and recovery horizons, predicted regret under the ERR-ES law closely matched observed post-failure regret measured from Monte Carlo rollouts, within delta less than or equal to 0.05. Our results reveal that recoverability is not an artifact of model scale or architecture, but a governed property of interaction dynamics, providing a theoretical foundation for execution-level robustness in language agents.
| Comments: | Preprint for ICML Submission |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2601.22352 [cs.LG] |
| (or arXiv:2601.22352v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2601.22352 arXiv-issued DOI via DataCite |
Submission history
From: Sri Vatsa Vuddanti [view email]
[v1]
Thu, 29 Jan 2026 21:55:50 UTC (155 KB)
[v2]
Fri, 2 Oct 2026 00:04:39 UTC (157 KB)
来源：arXiv：cs.LG · arxiv.org