Double-dimer model on a given graph is obtained by sampling two independent dimer configurations taken uniformly at random: this produces a number of loops and double edges, removing the latter one obtains what is called double dimer loop ensemble. Given a simply-connected domain on the complex plane consider its approximation by a sequence of Temperley domains on a square grid, where the step of the grid tends to zero. It was predicted by Kenyon that the corresponding sequence of double dimer loop ensembles converges to Conformal Loop Ensemble with parameter 4 (CLE(4)) sampled in the original domain. Recently this conjecture was supported by a breakthrough work of Dubedat: in this work a large family of observables, called topological correlators, was constructed and it was shown that their values for double-dimer loop ensembles converges to their values for CLE(4). Topological correlators carry a lot of topological information about loops by their construction and it was reasonable to expect that their values determine the probability measure on families of loops uniquely. As it turned out, values of topological correlators determine probabilities of a large class of events called cylindrical, which, in particular, implies the claim above: topological correlators do characterize a measure. In this talk we will discuss the construction of topological correlators and the machinery developed to extract concrete probabilities from their values. Based on a joint work with Dmitry Chelkak (Paris).