The most important general finding is that double dummy analysis is very accurate as compared to actual play from OKBridge. The overall total number of tricks taken by the declarer is 9.21 (9.22 for imp and 9.20 for mp). The double dummy analysis of the same deal produce 9.11 (9.12 for imp and 9.11 for mp). So actual play by OKBridge player takes 0.1 tricks more then the double dummy analysis result. This is from 383,000 deals and over 25 million plays.
This is the technique in solving how many tricks can be make for NT S H D C contracts when all 52 cards are exposed to all players. It assumes each player will make the best play possible when they select their cards to play. A double dummy database consists of 20 numbers for each bridge hand. The number of tricks the 4 declarers (north, south, east and west) can make in the 5 different types or denominations of contracts (spade, heart, diamond, club and no trump). From these 20 numbers the computer program can determine what the best score for the hand is, can a game or slam be made. The best score from the double dummy analysis is call par for the hand like in the game of gulf. This par value may not be a good number to use if you only look at one or two hands. If the only thing that is in question is which way to finesse a queen, the double dummy result will always find the correct direction. But on the other hand there are many hands the opening lead will give the declarer an extra trick which the double dummy defender will never make that “mistake”. If you use a large enough sample of bridge hands the double dummy par value for the hands are quite accurate and very realistic to be used in deciding what one should do. After studying hundreds of computer simulation results and thousands of hands manually, I notice the following small adjustment. The par value of double dummy analysis is a little bit higher then single dummy analysis for slam hands. This is due to more chances for two way finesse of queens and jacks and the dropping of singleton k and doubleton queens, and opening leads which do not give away a trick against slams is also a little bit easier to find. Currently I have a database of over 8 million randomly generated hands with double dummy analysis data calculated. This is good enough to do most research work on hand types that are common.
This presumably is going to be more accurate then double dummy analysis of bridge hands as compared to actual play of the hand by humans in a bridge tournament. Unfortunately as of now there is no large database of single dummy results of randomly generated hands. No software is fast enough to perform this task for millions of hands. Also there are other factors in single dummy analysis. The bidding by the defender give you extra information the single dummy analysis must use. This makes it very complicated to have a database of single dummy analysis of bridge hands.
Currently the only large amount of actual play of bridge hands are from the online OKbridge service. Between 1998 through 2001 there are about 383,000 deals and over 25 million plays in the database. The information is in public domain and you can buy a version of them from BridgeBrowser, which come with software that is targeted for individual player's record.
I have extracted deal, bidding, contract and result information from the OKBridge play records and added double dummy analysis information and created two databases. One is for extreme fast search and calculations which only store summary information for each deal. I have written a windows application which can search through the whole database and perform many statistical calculations in less then one minute. I also created a detail database which has information on each play of the deal. The speed on searching through this database is about 20 times slower. But the detail database has information on the individual bidding and the player names.