Faithful, Not Corrective: Message-Format Effects in Multi-Hop Agent Relays Are Tier-Dependent
Quick Answer
The study reveals that message format impacts multi-hop agent relays differently based on relay capability.
Quick Take
The study reveals that message format impacts multi-hop agent relays differently based on relay capability. Strong relays maintain nearly lossless fidelity across formats, while weak relays see an 8.7x increase in recall variability, influenced by encoding costs and schema resistance. Structure aids in error localization rather than correction, suggesting format choice should align with the weakest relay in the pipeline.
Key Points
- Strong relays show nearly lossless fidelity, with format-level fidelity unchanged under cognitive load.
- Weak relays experience an 8.7x increase in recall variability across formats.
- Rigid formats incur an encoding toll, impacting performance in weak relays.
- Injected errors persist to the final hop in 83-100% of chains, regardless of format.
- Structure improves error localization but does not serve as an error-correcting code.
Paper Resources
📖 Reader Mode
~2 min readAbstract:When LLM agents hand off information to one another, does the message format matter? Two literatures disagree: format-optimization work reports that structured messages cut cost without hurting accuracy, while format-restriction work finds that imposing structure degrades generation -- and neither measures what happens when a message traverses multiple hops, where copy fidelity, not one-shot generation, dominates. We introduce a controlled relay testbed: briefs of twelve programmatically generated atomic facts are re-encoded hop-by-hop in five formats (free NL, precision-instructed NL, JSON, triples, key-value) over six hops, scored by a fixed strong grader against programmatic ground truth, across two relay-capability tiers, a cognitive-load condition, and a paired-fork error injection. We find that message-format effects are tier-dependent. (i) Under faithful-relay instructions a strong relay is nearly lossless -- the documented "telephone-game" collapse does not occur -- and adding per-hop cognitive load leaves format-level fidelity unchanged (within +/-1.8 points) while raising generation cost by 24-53%. (ii) Under a weak (1.5B) relay the across-format spread of six-hop recall grows by a factor of 8.7 (from 2.3 to 20.5 points), driven by two opposing mechanisms -- an encoding toll paid by the rigid formats and drift resistance specific to the fixed-key JSON schema -- that flip the format ranking in transit. (iii) In a paired-fork injection, an injected wrong value, once present, persists to the final hop in 83-100% of chains in every format, closely matching each format's retention of the true value, with no detectable collateral damage to neighboring facts. Structure buys a faithful, error-localizing channel -- not an error-correcting code -- and format choice should follow the weakest relay in the pipeline.
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2607.09678 [cs.AI] |
| (or arXiv:2607.09678v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.09678 arXiv-issued DOI via DataCite |
Submission history
From: Zayx Shawn [view email]
[v1]
Fri, 12 Jun 2026 08:21:53 UTC (100 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Adversarial Social Epistemology for Assemblies of Humans and Large Language Models
The paper introduces Adversarial Social Epistemology (ASE) to analyze how agents manipulate trust in public communications, highlighting mechanisms that undermine the reliability of testimony and inference. It critiques existing frameworks like epistemic bubbles and misinformation diffusion, proposing a new language for understanding trust breaches and auditing inferential chains in densely interactive environments involving humans and large language models.