Introduction to Bridge Hand Simulation and Analysis

Computer reproduction of bridge hands yields amazing facts and insights into the way the cards work.  Not only can it be determined as to who can make what contract, but the data can tell us such things as whether it was statistically best to overcall on Board #13 from yesterday's duplicate.

Simulation data also can be utilized in the design of bidding systems.  For example: suppose that you were to analyze a few thousand deals in which opener held a balanced hand with 13-14 hcp and a 5-card major, and responder had a balanced hand with 3-card support and 10-11 hcp.  If the computer were, say, to inform you that major-suit game contracts rated to fail fully two-thirds of the time, would you not have cause to rethink your "automatic" limit raises with such holdings?

Now, imagine that the computer tells you that if your partner opens a 15-17 notrump and you are vulnerable in an imp-scored event, you would be way ahead in the overall total-points column simply by raising to 3 with every balanced 8-count!  Would you not then at least consider inviting game with some of those holdings?

These are the sorts of things that large-scale analysis can show us, and it is these findings that I will share on subsequent pages, with the aid of the double-dummy analyzer DeepFinesse, helpful input from local Sacramento player-programmer John Blubaugh, and Simulatron! — my ongoing programming project (SIM for short).  We are not interested in manipulating results, espousing a particular viewpoint, or in promoting any bidding system over another; we simply wish to know The Way It Is.

Admittedly, double-dummy results differ somewhat from those incurred at the bridge table.  Most notably, more defensive tricks are won by the computer than by real-life players.  The machine has the advantage of always knowing the best lead, always finding the missing queen, etc.  Conversely, at-the-table declarers have the advantage of winning a lot of extra tricks due to a preponderance of defensive errors.  This factor is huge, and it will be studied in detail.

Since I have found that data obtained from most runs of 1,000 hands do not differ significantly from samplings ten times that size, most studies will be in the range of 500 to 1,000 deals.

Typically, it takes my PC about 15 minutes to create 1,000 hands to specification and analyze them; additionally, human processing of the information and composition of web pages take a great deal longer.  As this project is important to me, I will do my best to publish regular postings.  Please bear with me, though, because — unlike some of my bridge partners — I do have a life on the outside.