Collective Counterfactual Planning: Coordination, Consent, and Verification under Representational Constraints

Details

Chainarong Amornbunchornvej introduces Collective Counterfactual Planning (CCP), a mathematical framework for when groups of agents can plan and execute goals exceeding any individual agent's capabilities — not due to knowledge or observability limits, but due to representational geometry. Each agent perceives the world through a projection onto a personal subspace of a shared task space. Four gates determine whether a team can reach a goal and legitimately certify completion: implementation coalitions (which agents must act together), plus three representational gates — conception, consent, and verification qualification. Two key results: First, iterated cross-agent relay can unlock solutions that no one-shot information pooling achieves — sequential mutual enabling lets teams exceed individual planning horizons. Second, any requirement hidden from the entire team's collective representation is unverifiable, even if the trajectory accidentally satisfies it. This framework formalizes why multi-agent AI systems may fail at verification even when succeeding at execution, with implications for multi-agent LLM orchestration, autonomous robot teams, and governance frameworks requiring certified goal completion.