The Cohen–Lenstra–Martinet heuristics are a probabilistic model for the behaviour of class groups of number fields in natural families. In this talk, I will discuss a generalisation to ray class groups. About 5 years ago, Varma determined the average number of 3-torsion elements in the ray class group of K with respect to m, when m is a fixed rational modulus, and K runs through the family of imaginary quadratic or of real quadratic fields. Since then, Bhargava has been challenging the community to come up with a natural probabilistic model that would explain the numbers obtained by Varma, and to predict more general averages in more general families of number fields. As I will explain in my talk, there turns out to be a very simple-minded way of doing so, and also a much more conceptual one, and they both turn out to be equivalent. The more conceptual one involves an object that does not appear to have been treated in the literature before, but that is very natural: the Aralelov ray class group of a number field. This is joint work with Carlo Pagano.