Why the equivalence problem for unambiguous grammars has not been solved back in 1966?

In 1966, Semenov, by using a technique based on power series, suggested an algorithm that tells apart the languages described by an unambiguous grammar and a DFA. At the first glance, it may appear that the algorithm can be easily modified to yield a full solution of the equivalence problem for unambiguous grammars. This article shows why this hunch is, in fact, incorrect.