In view of the hardly avoidable incompleteness of the recognizer coverage (see Section 1.4.3) we suggest to employ recognizer scores that are compatible with static evaluation scores. Hence, we base our recognizer scores mainly on the material balances of the according positions if neither draws nor mates get recognized.1.6 Compatibility with static evaluation scores has the further advantage of enabling the implementation to replace expensive evaluations by cheap recognitions. Simply skip the calls of the static evaluation function in case of successful recognitions and reuse the already computed recognizer scores as static evaluation scores.
In order to illustrate how scoring functions for interior-node recognition work, we develop such functions for the well-known classes KBBK (with differently coloured Bishops) and KBNK of elementary endgames. The winning strategy for KBBK with is to drive the lone King into any corner and mate it there by a concerted effort of all three pieces. The strong King must approach the scene, too, because two minor pieces cannot mate on their own. The same winning strategy applies to KBNK with the notable exception that the lone King has to be driven into a corner controlled by the Bishop of the strong side. Therefore, the distance between the lone King and the nearest target corner as well as the distance between the lone and the strong King are important traits of the respective positions. Beside these explicitly obvious relations of the two Kings the recognizer scores should also reflect the different degrees of difficulty involved in winning KBBK (with a maximal distance-to-mate of 19 moves) in contrast to winning the much harder KBNK (with a maximal distance-to-mate of 33 moves).
The below two scoring functions KBBK(p) and KBNK(p) implement the above guidelines for given positions p with all minor pieces valued at 3.25 Pawns and k corner dist(), kb corner dist(), as well as kk dist() suitably defined. Both scoring functions are efficient to compute and easy to implement. They let the search make progress in all recognized positions and channel it into greatly preferring KBBK over KBNK but not over positions where the winning side is up a Queen or more.
