We want to remove axiom layers from the translator output and instead compute
them in the search component where needed. This is clearer from a design
perspective since layers are not really a property of derived variables but
rather of an algorithm that computes their values. It also allows for different
layer strategies in different parts of the code.