Faithful, Not Corrective: Model Capability Governs Message-Format Effects in Multi-Hop Agent Relays
A controlled relay testbed is introduced in which briefs of twelve programmatic atomic facts are re-encoded hop by hop in five formats over six hops, scored against programmatic ground truth by a fixed strong grader, across two relay-capability tiers, a cognitive-load condition, and a paired-fork error injection.
Si-Cheng Zeng
· 0 citations