Independent of any implementation details, interior-node recognizers may be viewed as access functions to permanent databases of omniscient search results. In this rather abstract respect recognizers resemble transposition tables which also serve as databases of search results and are found in all decent chess programs. Therefore, one must integrate recognizers with transposition tables if planning to augment the search by interior-node recognition. But despite its obvious importance Slate's article does not mention anything about this task. Hence, we developed our own new scheme for the efficient yet seamless integration of recognizers and transposition tables. In the remainder of this section, we explain and discuss the peculiarities of our new scheme in more detail.
The central idea of our scheme is to handle interior-node recognition conceptually like an additional transposition-table access that gets executed after the normal ones. Thus, the search only falls back on interior-node recognition if all probes in the standard transposition tables fail. This greatly reduces the number of recognition trials whenever the rate of successful probes in the standard transposition tables is high (e.g. in endgames).
Treating interior-node recognition conceptually like an access to a transposition table provides for other advantages as well. In particular, our integration framework actually requires the results returned by recognizer calls to contain more information than just plain scores. Like entries in transposition tables they must at least comprise bound information, too. The following subsections discuss the peculiarities of our scheme in more detail.