Naïve averaging can approach truth at scale, but only while influence remains dispersed; homophily delays convergence, and a persistently dominant node can defeat it. This is a topology result, not an argument that consensus equals truth. For Cultural Intelligence, it makes dissent and attention architecture operational concerns; the architectural implication is to preserve enough independent influence for distributed evidence to survive aggregation.
Agents update by repeatedly averaging their neighbors' beliefs, the DeGroot rule. Nobody reasons about where the information came from. The question is whether a large society still converges on the truth.
It does, and the condition is about the network rather than the agents: convergence holds when influence stays dispersed, so no individual keeps a share of total influence that fails to vanish as the society grows. One sufficiently listened-to node is enough to break it.
Homophily slows convergence without preventing it. Groups that mostly listen to themselves take longer to agree with the world.