Some points in a random geometric graph cost no bits. An Evidence Press briefing. First, the background: scatter points at random in a region, and join any two that lie within a fixed distance of each other. Networks like this model wireless sensors, contacts and spatial infrastructure. A natural question is how many bits it takes, on average, to store such a network with n points. The rough answer is d times n times the logarithm of n, because each point has d coordinates, and the rest of the network pins each one down to about one part in n. Next, the question: in 2026, Vippathalla, Coon and Badiu proved upper bounds for the cube and the torus, and conjectured that their leading terms were exactly right. Now, the main finding: on any convex region, and for every fixed distance, the leading cost is d times the fraction of points that are not universal, times n log n. A universal point is within reach of every possible position, so it is joined to everyone, and its links reveal nothing about where it is. Such points cost nothing. In the unit square at distance one, about thirty-one and a half per cent of points are universal, and the true coefficient is about 1.370, not the conjectured 1.461. So the exact cube conjecture is false at every radius between half the diagonal and the full diagonal, in two or more dimensions. It is confirmed elsewhere: for smaller radii, in one dimension, and on the torus at every radius, where every direction is visible. For more general connection rules on the torus, the coefficient counts the directions the rule can detect. As for how sure this is: every theorem has a written proof, and the computations are sanity checks. Finally, the limits: the results are for a fixed distance as n grows, and they do not identify the next term. It is an anonymous, unrefereed candidate, with AI-assisted review only, not peer review. Published on 27 October 2026, its paper and evidence are linked from the Evidence Press release page. This briefing uses an AI-generated voice and is not additional scientific evidence.