The use of an opening database in an engine vs engine match has some interesting applications. For instance, you can start a theme tournament to test the quality of your favourite openings system, store various lines in the database and then have the program play them out against itself. Or, you can store a series of games or positions you want to analyse in a database and start the program working on them. It can also be used to conduct the well-known “Nunn test”.

 

GM John Nunn has selected a series of ten openings to test engine strength. The opening database required to conduct the match is to be found on your program CD, and is automatically installed on your hard disk (in the directory \testsets). If you start a match between two engines using this “Opening DB” and “Alternate colours”, you will get 20 games which give you a very good idea of the pure playing strength of the engines.

 

There are a number of reasons for this kind of experimental match, which eliminates the use of traditional openings books:

 

   With the amount of effort going into the production of optimised openings books, and with the advent of learning books, the measurement of the playing strength of the chess engine itself has been pushed into the background. A large number of computer vs computer games are decided by the openings book.

 

   It is possible to improve the results of one program against a competitor by simply “booking” the opponent (i.e., including lines in the books that are aimed at a specific program). Such “killer openings” appear repeatedly on the boards of testers and drastically improve the score of a program, without need for improvement in the playing strength of its engine.

 

   It is important to define a fixed set of games that should be played in a test, rather than to use a random set of games, where there is a danger of selective reporting. A Nunn match is only valid if exactly twenty games are submitted. It is entirely reproducible and transparent to everyone. This is even more true if the precise hardware configuration is known, and the thinking times and evaluation of each move recorded, which the program does automatically.