Bayes-Sufficient Compression Is Not Enough: How Does Communication Help Multi-Agent Systems?
Abstract
Multi-agent LLM systems pair a sender with broad context and an executor with a limited local view. We study when a short message improves the executor's next decision, when raw context is preferable, and when a stronger sender helps. Our framework, \emph{receiver-relative bounded coordination}, expresses message utility as receiver gain minus protocol tax. Compression beats raw context when tax savings exceed losses from omitted information and decoder mismatch. Even \emph{Bayes-sufficient} compression can fail when a bounded executor cannot use its surface form. A three-stage decomposition separates externalization, absorption, and \emph{action closure}, explaining how errors remain after the correct content reaches the receiver. Under a single-crossing condition, sender upgrades help above a receiver-burden threshold. Across six benchmarks, the same Qwen protocol raises ContextBench joint accuracy from $0.633$ to $0.775$ but lowers ToolSandbox from $0.889$ to $0.653$. Fixed-message replay reveals closure failures despite correct artifact recovery. These results guide an inference-time selector that improves the accuracy-cost frontier on the evaluated communication regimes.
Related
- Necessary and sufficient conditions for optimal control of coupled forward-backward stochastic control systems driven by G-Brownian motion
- Necessary and Sufficient Conditions for Adaptive Stabilizability of Continuous-Time Markov Jump Linear Stochastic Systems
- When Is Enough Enough in Self-Evolving LLM Systems?
- Classical Communication Helps Single-Shot Quantum Compression
- Dual-Branch Vector-Quantization-Aided Satellite Digital Semantic Communication with Index Compression for High-Resolution RSI Over AFDM
BibTeX
@misc{xie2026bayessufficient,
title = {{Bayes-Sufficient Compression Is Not Enough: How Does Communication Help Multi-Agent Systems?}},
author = {Yi Xie and Zhanke Zhou and Yi Fan and Yong Ge and Bo Han and Bo Liu},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.03769},
eprint = {2610.03769},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.03769},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Bayes-Sufficient Compression Is Not Enough: How Does Communication Help Multi-Agent Systems?
AU - Yi Xie
AU - Zhanke Zhou
AU - Yi Fan
AU - Yong Ge
AU - Bo Han
AU - Bo Liu
PY - 2026
DA - 2026/10/01
PB - arXiv
DO - 10.48550/arxiv.2610.03769
UR - https://arxiv.org/abs/2610.03769
AB - Multi-agent LLM systems pair a sender with broad context and an executor with a limited local view. We study when a short message improves the executor's next decision, when raw context is preferable, and when a stronger sender helps. Our framework, \emph{receiver-relative bounded coordination}, expresses message utility as receiver gain minus protocol tax. Compression beats raw context when tax savings exceed losses from omitted information and decoder mismatch. Even \emph{Bayes-sufficient} compression can fail when a bounded executor cannot use its surface form. A three-stage decomposition separates externalization, absorption, and \emph{action closure}, explaining how errors remain after the correct content reaches the receiver. Under a single-crossing condition, sender upgrades help above a receiver-burden threshold. Across six benchmarks, the same Qwen protocol raises ContextBench joint accuracy from $0.633$ to $0.775$ but lowers ToolSandbox from $0.889$ to $0.653$. Fixed-message replay reveals closure failures despite correct artifact recovery. These results guide an inference-time selector that improves the accuracy-cost frontier on the evaluated communication regimes.
ER -