A graph of two hundred people in a clustered slab, each joined to about ten others they can see, with a few links that go a long way. The graph sits dark; two people are seeded with a claim apiece, at random and far apart; each convinces a couple of immediate neighbours, because a single carrier can never reach a threshold of two. After that it is only the rule. A person commits the moment enough of the people they can see hold a claim, but does not hold it, and cannot pass it on, until it has crossed the distance to them — so what is drawn is a front advancing one person at a time, not a graph repainting. A control sets the threshold. Measured over twelve runs at each setting: at one, everything floods; at two, about four fifths of the network decides over roughly twelve seconds, with the leading claim taking a little over half of everyone and the trailing one about a quarter; at three, under three fifths decide; at four and above nothing leaves the seeds. Where the two fronts meet they stop each other, and the people caught between them never decide at all — that residue is not a failure of the model, it is most of what the model is for. Which claim leads is settled in the first handful of hops by where the seeds happened to land relative to the dense clusters, and re-running with the same threshold reverses it about as often as not. Afterwards the leading claim is held by more people than anything else in the network, is not distinguishable from knowledge by any test available inside it, and the other is a minority position that would, if pressed, look like a symptom.