# Context Is a Pipeline, Not a Window: Information Attrition in LLM Agent Harnesses

arXiv:2610.36114v1 Announce Type: cross Abstract: LLM agents are judged by two numbers: the size of the context window, and how often the final answer is correct. Both numbers miss a third thing: how much of a tool's output is lost before the model ever sees it. We measured this loss in a real tool-using agent. The agent finds the cause of injected faults by reading Linux kernel traces. The context window was almost empty: across 720 runs, the single largest prompt used only 5.75% of it. Yet a third of all tool results (3,827 of 11,189) were cut before they reached the model, and every cut result became broken JSON. The harness cut each result by slicing its JSON text at a fixed length. This does not shorten a result evenly. It deletes whatever data sits near the end. One field reached the model in only 4.7% of calls. A second field, in the very same results, reached the model in 99.3% of calls. Which field held the fault's evidence depended on when the faulty process started: early or late in the trace. The agent's diagnosis accuracy matched this same split. So we ran controlled experiments to test one question: did the data loss cause the drop in accuracy? This paper makes four contributions. First, we model an agent's context as a five-stage pipeline, with a measured failure at each stage. Second, we replay 864 real tool results, including one from a production trace where the shipped cut deletes all 128 records of the fault's evidence. Third, a controlled experiment shows that the cut's real cost appears first as extra work, not as wrong answers: 43% more tool calls, twice the computation, and yield falling from 0.61 to 0.31. The agent still finds the right answer when it has another way to reach it. Fourth, a survey of seven agent frameworks finds two that cut tool output the same unsafe way. Whether the data cut truly causes the accuracy drop is still an open question.

## Why It Matters

This paper models an LLM agent's context as a five-stage pipeline and measures how much tool output is lost before the model ever sees it. In a real tool-using agent, a third of all tool results were cut before reaching the model and every cut became broken JSON, and a controlled replay showed the real cost appears as extra work—43% more tool calls and twice the computation—rather than immediate wrong answers. A survey of seven agent frameworks found two cut tool output the same unsafe way. This matters because context-window size alone misses the attrition that harms real agents. See [Agents, Autonomy, and Tool Use](../agents-autonomy.md), [Agent Tooling](../agents-tooling.md), [Large Language Models](../large-language-models.md), and [Inference Serving](../inference-serving.md).

## Sources

- rss: https://arxiv.org/abs/2610.36114

*Back to [Recent News](../../news.md) · [Sub-article index](../index.md).*