Fidelity Collapses with Table Size, Not Turn Count: Delivery Architecture in a Deployed Tool-Using Agent
Factual evidence
What the source reports
Research shows multi-turn LLM agents corrupt structured data based on context table size rather than turn count during context re-emission.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 4 November 2026
- Collected by OneBench
- 5 Nov 2026, 02:01 UK
Stored source excerpt
arXiv:2611.00442v1 Announce Type: new 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…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Passing complex structured data through multi-turn agent contexts risks data corruption that scales directly with payload table size.
Do what
Review agent architecture choices with technical leads to avoid relying on model context for multi-turn structured data retention.