arXiv：cs.LG· Yue Zhao·· 4 小时前AI 评分41
GRADE：用图结构表示 LLM 智能体的依赖与执行
GRADE: Graph Representation of LLM Agent Dependency and Execution
AI 导读
GRADE 将 LLM 智能体的执行轨迹与依赖关系统一表示为一张图，已发表的六种图视图均为其投影。执行边可直接获取，依赖边按“轨迹中观测、新增日志声明、假设下推断”分为不同等级，研究从六个工具调用、编程和网页语料中恢复依赖层，并检验其对运行规模的失败预测增益：在三个语料上优于线性运行规模基线，但增益取决于学习器和任务采样；在留出语料上该依赖层仍高于随机水平。
正文
Abstract:Traces record what LLM agents did, leaving implicit what each step relied on. GRADE holds both in one graph; six published graph views are its projections. Execution edges come free; each costly dependency edge is graded as observed in the trace, declared by added logging, or inferred under an assumption, a schema we test one grade at a time. We recover the dependency layer from six tool-use, coding, and web corpora and measure its failure-prediction gain over run size. First, over a linear run-size baseline, it improves prediction on three corpora, but the gain depends on the learner and task sampling. On held-out corpora the layer stays above chance where run size does not, under one learner, with the same task-sampling caveat. Second, structure can pass for signal it does not carry: a layer inferred under full history restates run size. Within a corpus, a preregistered test does not separate the recovered layer from a degree-matched counterfeit that rewires each step's dependencies but keeps their number. The recovered layer's largest transfer lead over it reverses when the held-out corpus's task family is withheld. Before crediting structure, grade each edge, keep a run-size baseline, and score degree-matched attachment controls.
| Comments: | 35 pages. Expanded analyses and revised figures. Code: this https URL |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.22741 [cs.LG] |
| (or arXiv:2606.22741v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.22741 arXiv-issued DOI via DataCite |
Submission history
From: Yue Zhao [view email]
[v1]
Mon, 22 Jun 2026 01:03:21 UTC (507 KB)
[v2]
Sat, 12 Sep 2026 08:01:55 UTC (584 KB)
[v3]
Tue, 6 Oct 2026 22:44:44 UTC (958 KB)
来源：arXiv：cs.LG · arxiv.org