生成式评估套件作为多样性驱动的搜索:检测真实 Agent 回归
Generated Evaluation Suites as Diversity-Driven Search for Real Agent Regressions
AI 导读
arXiv 论文提出将智能体评估套件生成视为多样性驱动的搜索,并向 τ²-bench 智能体注入 7 个受控缺陷来检验缺陷检测能力。研究发现缺陷可分为迹对齐与迹正交两类:对齐类中逐格打分可完整恢复工具移除缺陷;仅含 happy-path 的套件检出率为零,而 kind 多样化的套件在 20 次中 5 次检出编排缺陷。论文建议评估档案按单元格打分,并将覆盖率作为一等属性发布。
正文 · 原文
Abstract:Generating an evaluation suite from an agent's specification is diversity-driven search: a generator fills a descriptor space (requirements $\times$ personas $\times$ scenario kinds), and the suite it emits is an archive, then scored like a benchmark by its mean pass rate, which throws away the search's product. We inject 7 controlled defects into a $\tau^2$-bench agent across two domains. Coverage yields a ceiling: if a defect's traced footprint holds $b$ scenarios that pass at baseline, it can itself contribute at most $b/N$ to a suite of $N$. For 4 of 5 traceable defects that ceiling lies at or below the detection threshold $\theta = 0.10$, so no within-footprint effect can move the suite past it; in both domains the tool-removal defect destroys every covered scenario it can (exactly its ceiling) while the suite-level pass rate moves the other way. Defects split into two regimes: trace-aligned ones, where per-cell scoring recovers the tool-removal defect in full and reveals a second as a significant improvement, and trace-orthogonal ones, whose footprint is behavioural: forced escalation is detected in retail at a traced coverage of exactly zero. Diversity along the generator's own axes helps only where it aligns: at fixed suite size a happy-path-only suite detects nothing, while a kind-diverse one detects the orchestration defect in 5 of 20 draws. Evaluation archives should be scored per cell and should publish coverage as a first-class property.
| Comments: | 9 pages. Accepted at the NeurIPS 2026 EvoRobust Workshop |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.17632 [cs.AI] |
| (or arXiv:2611.17632v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.17632 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Pulkit Yadav [view email]
[v1]
Fri, 13 Nov 2026 01:03:19 UTC (22 KB)
来源:arXiv cs.AI · arxiv.org