Mike S. wrote:I can only imagine that such blunders come into a book by importing a "broken" game. I guess almost all big databases to work with, will contain a couple of such games.
But normally, an Arena book will never play such moves if the book option
Minimal number of games: 2
(or bigger; Fritz' "optimal setting" value is 3 here)
because it is highly unlikely that moves like 6...Ba3 and 9...Qc6 were played more than once. - The min. percentages should be another safety net against such moves, although if a game was broken the result could be wrong, too.
So, even if a book or the database it is based upon, does contain some of such blunders, proper book options should prevent that they are played from an Arena book.
In making an opening book there are certain precautions that should be taken.
A book is derived from the positions in a database of played games. The quality of the games in the database can be very high overall but there are bound to be a few undetected games containing tactical blunders of varying depths (1 move, 2 move, etc).
This is an understandable occurrence with human players. The best way to eliminate the tactical blunders is to do a computer analysis of each move of each game of the database. Since the database is usually very large, that is not practical. That brings up the issue of "filtering" out potential blunders from the database as the opening book is created or being used.
I recommend 5 types of filtering.
First, you have to have games between expert players. For human games I would only use games where both players have an elo of 2400+. Games between chess engines are dangerous to use in creating opening books. Although the elite chess engines do not make short tactical blunders, they themselves use opening books that may contain blunders in the first 12 moves, as shown by the example that started this thread.
Secondly, I would require a minimum number of position occurrences beyond "1". That means that if a tactical blunder occurs, it had to have been made more than once. I prefer using "3" as the minimum. However a higher minimum has a cost. A brilliant move that occurred less often than the minimum would not be included in the book.
Thirdly, I would require all games in the database to have lasted substantially beyond the opening. If I am making an opening book of 12 moves, I would only use a database of games that lasted beyond 24 moves. The reasoning is that a game that ended early might have terminated because of an opening blunder. Another reason to require longer games is to filter out "grandmaster draws" which are games drawn for personal reasons even if one of the players had a substantial advantage in the opening.
Fourthly, I would only use games in the database played within the last 10 years. Opening theory is always evolving, especially with computer analysis.
Fifthly, I would only use games in the database that were played with full time-control, as opposed to blitz and rapid games.
By the way, I have created databases of games meeting the criteria I have mentioned. See my home page (click on "www" below). The databases are separated into different time spans.