Is this about the problem that Prolog tends to ping pong infinitely between leafs? I wrote a fully declarative solitaire game solver and remember this being a big issue forcing me to memorize past states of the backtracking and basically exclude them by another predicate. This is obviously slow. I thought, why not at least have a plugin to avoid trivial cases where backtracking gets stuck switching between A and B. Or a plugin for a stochastic traversal of the solution tree.