Swapping a Role-Matched Agent Costs Little Task Score but Raises Communication Per Unit of Progress 16-63%

arXiv 2609.05279·medium signalXBlueskyLinkedInCopy link

arXiv 2609.05279 tests the production assumption that an agent filling a role is interchangeable with any other agent that can do the job. Eight teams per setting form independently from one base model, each agent keeping a private notebook across ten formation episodes, then role-matched agents are traded between teams; against a placebo reproducing roster-change disruption without changing the occupant, a swap barely moves task score but raises communication spent per unit of progress by 16 to 63 percent. In Hanabi a swapped agent is more expensive than an inexperienced one, and in Collab-Overcooked most of the extra communication comes from the agent that stayed. Greedy decoding lowers the penalty; doubling formation history raises it.