21.3 Run MINIHYPER with the predecessor exercise and measure the execution time. How many hypotheses are generated, how many are refined? The 'guard' literal atom(X) in the background literals above is helpful, but there is a problem. At the moment that a literal pair like [atom(X),parent(X,Y)] is added, all the goals to it will fail (because X is a new variable not yet matched to anything else). Only after a further refinement, when X is matched to an existing variable, may such a literal succeed. So after a refinement with matching, such a hypothesis may become more general. This is different from the usual case when refinements produce more specific hypotheses. This anomaly in the case of atom(X) makes the refinement process non-monotonic: the hypotheses along a refinement path are not necessarily increasingly more specific. In this case an incomplete hypothesis may become complete after a refinement which violates the basic assumption on which the search of the hypothesis space is based! However, in this particular case the search still works. Find a sequence of refinements of complete hypotheses in the refinement graph that, despite this anomaly, leads to the target hypothesis.