[kenberg]

We are told that Gib uses double dummy simulations to choose it's defense. Fine, but I am hoping that we can explore this a bit.

Suppose Gib is on lead in the middle of a hand and the lead will clearly be either A or B.Suppose simulation says that 100% of the time the A lead will allow declarer to make x tricks.Suppose simulation also shows that the B lead will hold declarer to x-1 tricks sometimes, to x tricks other times and give him x+1 tricks other times.
Humans encounter this problem and do their best. What do the Gibs do?
Perhaps the level of the contract matters? If declarer is in 4S and x=10, does Gib go all out to beat the contract by choosing the B lead even though it might give declarer an overtrick? Maybe it depend on how likely the overtrick is?

A recent example, sort of.
http://tinyurl.com/y8y3d5ot
At T7 LHB led his now bare club Q. Thanks buddy. Maybe I was going to drop it anyway, maybe I wasn't. A double dummy evaluation of course says I would. Let's say probably I would. I still appreciated the bot's effort on my behalf

It's true that if RHB has the club K then I will go down after the Q lead but these simulations are supposed to take the bidding into account. RHB had AKJxxx in diamonds and opened 2D. He doesn't have the club K. Now at least by dd, he did not give me a trick. Suppose he plays his last spade. I clearly have four spades, three hearts, two clubs so I have my contract. When I get to leading the club I will know that RHB started with three. But Qxx seems possible even with the AKJ of diamonds. Maybe i get it right, maybe wrong.

This example isn't perfect, but over time I have seen quite a few cases where I am not sure what simulation led to that play.

I just find this interesting. I am not all that much planning on telling the programmers what they are to do, I just enjoy trying to get inside the bot brain. Why the club Q? In a similar situation LHB, in the middle of a hand, led his stiff K of hearts with QJx on the board. Yes I had Axx. Double dummy I can drop the K, got that, but still it was a strange lead. That one was too long ago for me to locate.

[Stephen Tu]

These are just classic double dummy reasoning flaws. GIB assumes DD declarer play. So it gives up, it assumes all cards are equally hopeless and sometimes randomly leads its stiff honor. The solution is for GIB to simulate presuming *single dummy* declarer play, then it would realize that not playing the stiff honor might lead to declarer getting it wrong sometimes and score leading the honor lower than alternatives. But this takes way too much time.