← 研踪 PaperTrailarXiv:2611.17632 · v1
Generated Evaluation Suites as Diversity-Driven Search for Real Agent Regressions
首次提交 2026-11-13 09:03
版本更新 2026-11-13 09:03
雷达观测 2026-11-17 04:57
摘要
Generating an evaluation suite from an agent's specification is diversity-driven search: a generator fills a descriptor space (requirements personas 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 -bench agent across two domains. Coverage yields a ceiling: if a defect's traced footprint holds scenarios that pass at baseline, it can itself contribute at most to a suite of . For 4 of 5 traceable defects that ceiling lies at or below the detection threshold , 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.
作者备注:9 pages. Accepted at the NeurIPS 2026 EvoRobust Workshop
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-13 09:03 · 最新观测版本
Generated Evaluation Suites as Diversity-Driven Search for Real Agent Regressions
Generating an evaluation suite from an agent's specification is diversity-driven search: a generator fills a descriptor space (requirements personas 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 -bench agent across two domains. Coverage yields a ceiling: if a defect's traced footprint holds scenarios that pass at baseline, it can itself contribute at most to a suite of . For 4 of 5 traceable defects that ceiling lies at or below the detection threshold , 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.
查看此版本解读 →查看此版本原文 ↗