a rumour does not need everyone to hear it in order to die — it needs enough people to have heard it that telling stops working
R803’s engine needed a difference to work across. Here is a thing that works by levelling the difference it feeds on. A rumour lives in a gap: some people know, and some do not. Telling closes the gap, one hearer at a time. But closing it does two things at once. Every new knower is one fewer person left to tell — and one more person a teller can meet who already knows, which is exactly the meeting that makes a teller stop.
Daley and Kendall modelled this in 1964 (“Epidemics and rumours”, Nature) and 1965 (“Stochastic rumours”). Split a closed population into three parts. Ignorants have not heard it. Spreaders have heard it and are passing it on. Stiflers have heard it and have stopped. Two kinds of meeting drive everything:
— a spreader meets an ignorant: the ignorant becomes
a spreader. The gap narrows from the ignorant side.
— a spreader meets anyone who already knows (another
spreader, or a stifler): the spreader becomes a stifler.
The gap narrows from the spreader side too — in the
Daley–Kendall version, a spreader–spreader meeting
stifles both.
Early on, almost every meeting is of the first kind, so spreaders multiply. Late, almost every meeting is of the second kind, so spreaders drain away faster than they are made. Nothing outside has to intervene. The rumour spends its own difference, and the levelling is its whole life: it succeeds by producing the conditions in which it can no longer succeed.
Run the levelling. The box follows the deterministic Daley–Kendall equations for the three fractions — ignorant x, spreader y, stifler z, with x + y + z = 1: x′ = −xy, y′ = y(2x − 1), z′ = y(1 − x). It starts with one spreader in a thousand.
And here is the part that corrects the easy version of the story. The rumour does not die by being believed by everyone. When the last spreader stops, in the large-population limit of both the Daley–Kendall model and the later Maki–Thompson variant (1973), about 20.3% of the population has still never heard it at all — x∞ ≈ 0.2032, the root of ln x = 2(x − 1) besides x = 1. The levelling stops before it is complete, because levelling was never the goal. There is no goal. There is only a process whose fuel and whose brake are the same meetings, and the brake wins first.
Three narrowings. First, this is a closed, well-mixed population: everyone meets everyone else with equal chance, nobody forgets, nobody arrives, and there is no broadcast — no announcement that reaches the remaining fifth at once. Real rumours run on networks, in clusters, with hubs and outsiders, and their final reach differs. Second, the 20.3% figure is a limit result for a large population under these rules, not a prediction about any actual rumour. Third, the model is indifferent to truth: the rumour may be true or false, and the maths does not change. What kills it is not refutation. It is saturation, arriving early.