As AI agents become more tightly linked through networks, many tasks require negotiation and joint decision‑making. To address this, we introduce an evaluation framework that treats consensus formation as the evolution of publicly stated positions toward a shared region, using the accumulated communication history as context. The framework records each agent’s position trajectory, group dispersion, and the content expressed in public messages.\ \ We validate the framework with two case studies. The first involves staffing decisions based on distributed facts; the second requires agents to reconcile duplicate information and a cancelled contract. Across 1,872 completed runs, sharing explanations correlates with higher group decision accuracy, while the second task shows persistent disagreement despite correct aggregate answers. Membership‑change experiments further demonstrate the framework’s applicability in dynamic networks. By separating agents’ convergence from progress toward the task’s full‑information answer, the framework links decision accuracy and bounded disagreement to the information available during exchange.\ \ Review