arXiv cs.CL· Tung Thanh Hoang·· 12 小时前AI 评分58
论文：工具型智能体的数据保真度随表格规模崩塌，而非对话轮数
Fidelity Collapses with Table Size, Not Turn Count: Delivery Architecture in a Deployed Tool-Using Agent
AI 导读
一项在生产级差旅智能体上的对照实验显示，把工具返回的结构化记录直接放进规划器上下文并在每轮重新输出（SYNTHESIS），24 段对话中仅 45.7% 的轮次完全正确，而把记录留在会话状态、经确定性过滤后通过类型化旁路通道交付（BYPASS）达到 92.3%，配对差距 +46.6pp。
正文 · 原文
Abstract:When a multi-turn agent delivers a tool's structured records into the model's context and re-emits them each turn, every turn can corrupt data the tool computed correctly. We measure this in a production award-travel agent: two architectures over identical frozen payloads, lineage-matched prompts and sampling, differing only in delivery architecture. SYNTHESIS feeds full record sets into the planner's context and answers follow-ups in-head; BYPASS, the deployed system, keeps records in session state, translates follow-ups into arguments for a deterministic filter, and delivers rows over a typed side channel. Across 24 dialogues (339 scored turns per condition), BYPASS scores 92.3% of turns fully correct, SYNTHESIS 45.7% (paired gap +46.6pp [+35.7, +57.1]). Varying delivered-table size by capture date, SYNTHESIS collapses from near-perfect at 10 rows or fewer to 8% or less beyond about 74 rows. On a content-controlled prefix ladder the collapse threshold orders by model capability, from under 10 rows (GPT-4o-mini) to about 198 or more (Gemini 2.5 Pro), while BYPASS stays size-flat for all four models. The collapse is expensive: beyond 100 rows, where almost no SYNTHESIS answer is correct, each correct one costs roughly 200x BYPASS's flat \$0.012. Practitioners converged on typed side-channel delivery for rendering, not correctness; we show it was, unknowingly, also a correctness decision.
| Comments: | Non-archival paper at the Second Workshop for Research on Agent Language Models (REALM) at EMNLP 2026. 21 pages, 5 figures |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.00442 [cs.CL] |
| (or arXiv:2611.00442v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.00442 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Tung Hoang [view email]
[v1]
Fri, 30 Oct 2026 22:50:32 UTC (142 KB)
来源：arXiv cs.CL · arxiv.org