arXiv：cs.AI· Fangfei Li, Chenyang Zhao, Long Wang, Feng Tian, Lv Guo·· 4 小时前
Trace Evolution Bench（TEB）：衡量智能体能否从自身环境轨迹中进化
Trace Evolution Bench: Measuring Whether Agents Improve from Their Own Environment Traces
AI 导读
研究者提出 Trace Evolution Bench（TEB），将冻结重评分与上下文或权重更新区分开，并配套三个冻结的 80/20/40 数据包，以隐藏单元测试评分的编码任务为主榜。
正文
Abstract:Self-evolving agents are usually scored by endpoint Gain: held-out success after an update minus success before. That difference can credit a frozen rescore as evolution. We introduce Trace Evolution Bench (TEB), a protocol that separates that rescore from a context or weight update, together with three frozen 80/20/40 packs. An update is accepted only when validation does not fall, held-out identifiers never enter the update, and a text update is not ranked with a weight update. The primary scoreboard is coding, graded by hidden unit tests in a closed workspace. On that pack (40 held-out tasks, one family), DeepSeek v4 Pro under no_evolve never changes its policy yet moves 7.5, 5.0, and 12.5 percentage points across three seeds. Of nine text-update cells inspired by GEPA, ACE, and ExpeL, seven have positive excess above that seed's own band, and one gepa_llm seed moves down by the width of its band (-12.5, regression 0.48). GLM and MiniMax do not reproduce that seven-of-nine pattern. No method stays above its own band on every seed of all three models: GLM has two paired cells above their own bands, and MiniMax expel_llm seed 44 is a third (regression 0.30). The methods were fixed on a validation screen before held-out was read. Office, graded by rule-bounded delivery, and manufacturing table QA are separate packs: office DeepSeek does not exceed its own bands (widest 17.5 points), and manufacturing has no paired positive excess beyond its seed band.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15055 [cs.AI] |
| (or arXiv:2610.15055v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15055 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Fangfei Li [view email]
[v1]
Sat, 10 Oct 2026 16:45:52 UTC (298 KB)
来源：arXiv：cs.AI · arxiv.org