arXiv cs.AI· Veljko Skarich·· 2 天前AI 评分29
Poison Valleys：诊断 LLM 代码修复中的路径依赖失败
Poison Valleys: Diagnosing Path-Dependent Failure in LLM Code Repair
AI 导读
Poison Valleys 在 1,844 次模型请求中追踪 LLM 代码修复的路径依赖失败，覆盖 PersistQueue 与 Dijkstra 两个合成 Python 案例。将错误机制从假设移入实现计划后，PersistQueue 失败率上升 32.2%，Dijkstra 上升 28.9%。不同误解产生模型相关的义务违反模式，但可能收敛为相同行为。
正文 · 原文
Abstract:When code-repair agents fail, final tests reveal that a patch is wrong but not which earlier commitment caused it or which software obligation failed. Poison Valleys traces this path in 1,844 model requests on two synthetic Python cases: PersistQueue adds atomic batch claiming to a disk-backed queue, while Dijkstra adds source-sensitive filtering to shortest-path search. We define plausible but incorrect software mechanisms, place each in a model's hypothesis, implementation plan, or both, and evaluate the generated code with an Oracle and targeted diagnostic coordinates. Moving the misconception from hypothesis to plan increased failure by 32.2% in PersistQueue and 28.9% in Dijkstra after matching the experimental mix. Different misconceptions produced distinct, model-dependent patterns of violated obligations, although several could converge on the same behavior. This creates a measured connection from upstream misconception to code mechanism and failure form. Code, data, and demo: this https URL.
| Comments: | Accepted to Interpreting Agent Behavior (IAB): The First Workshop on Human-Centered Interpretation for Understanding Agents, Humans, and Interaction (NeurIPS 2026 workshop); 22 pages; code, data, and demo: this https URL |
| Subjects: | Software Engineering (cs.SE); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.21514 [cs.SE] |
| (or arXiv:2610.21514v1 [cs.SE] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21514 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Veljko Skarich [view email]
[v1]
Thu, 15 Oct 2026 08:49:49 UTC (37 KB)
来源：arXiv cs.AI · arxiv.org