Faithful, Not Corrective: Message-Format Effects in Multi-Hop Agent Relays Are Tier-Dependent
Faithful, Not Corrective: Message-Format Effects in Multi-Hop Agent Relays Are Tier-Dependent
忠实而非纠错:多跳智能体中继中的消息格式效应具有层级依赖性
Abstract: 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.
摘要: 当大语言模型(LLM)智能体之间传递信息时,消息格式是否重要?目前学术界存在两种不同的观点:格式优化研究指出,结构化消息可以在不损害准确性的前提下降低成本;而格式限制研究则发现,强制使用结构会降低生成质量。然而,这两种研究都没有衡量当消息经过多跳传递时会发生什么,在多跳场景下,复制的忠实度而非单次生成能力起着决定性作用。
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.
我们引入了一个受控的中继测试平台:将十二个程序生成的原子事实摘要,通过六个跳数,以五种格式(自由自然语言、精确指令自然语言、JSON、三元组、键值对)进行逐跳重新编码。测试涵盖了两个中继能力层级、一个认知负荷条件以及一个配对分支错误注入,并由一个固定的强评估器根据程序生成的基准事实进行评分。
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%.
研究发现,消息格式的效应具有层级依赖性。(i)在忠实中继指令下,强模型中继几乎是无损的——文献中记载的“传话游戏”崩溃现象并未发生——并且增加每跳的认知负荷不会改变格式层面的忠实度(波动在 +/-1.8 分以内),但会使生成成本增加 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.
(ii)在弱模型(1.5B 参数)中继下,六跳召回率在不同格式间的差异扩大了 8.7 倍(从 2.3 分增加到 20.5 分)。这是由两种相反的机制驱动的:刚性格式需要支付的“编码代价”,以及固定键 JSON 模式特有的“漂移阻力”,这两种机制导致了格式排名在传输过程中的反转。
(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.
(iii)在配对分支注入测试中,一旦注入错误值,在所有格式的 83-100% 的链条中,该错误都会持续到最后一跳,这与各格式对真实值的保留率高度吻合,且对相邻事实没有产生可检测到的附带损害。结构化提供的是一个忠实的、错误定位的通道,而非纠错码;因此,格式选择应遵循流水线中最弱的中继节点。