Auditing Multi Agent Influence Controls: Action Constraints and Tests for Useful Decomposition

This paper develops an audit specification for separately controlling participation, signal gain, timing, information exposure, and permission in multi-agent workflows. Dynamic Influence Balance names the candidate architecture. The analysis separates observations, actuators, authorization, and protected outcomes, and examines restricted action directions, nonnested decision regions, joint constraint feasibility, and limits of independent ablation results. Revised P70-P75 questions and a complete-system comparison test whether the decomposition improves useful service after coordination costs. An earlier implementation summary is excluded because execution records and its generator are unavailable. The architecture's incremental benefit, minimality, and deployment safety remain unvalidated.