Asymptotic extremal combinatorics deals with questions that in the language of model theory can be re-stated as follows. For finite models M, N of an universal theory without constants and function symbols (like graphs, digraphs or hypergraphs), let p(M, N) be the probability that a randomly chosen sub-model of N with |M| elements is isomorphic to M. Which asymptotic relations exist between the quantities p(M₁, N),...,p(Mh, N), where M₁,...,Mh are fixed "template" models and |N| grows to infinity? In this paper (...) we develop a formal calculus that captures many standard arguments in the area, both previously known and apparently new. We give the first application of this formalism by presenting a new simple proof of a result by Fisher about the minimal possible density of triangles in a graph with given edge density. (shrink)