[BunnyGo]

This hand was played in an automated free individual event. The clock ran out with 2.5 tricks to play, and was resolved in favor of declarer guessing correctly:

---

Note that if West pitches the diamond 8, then North has to guess if hearts are 1-1 or 2-0 at the end.

Now that I look at the hand more closely, I realize that there is no misguessing (as pitching the spade 4 and finessing, or ruffing the Ten and playing for 1-1 both work). However, I'm interested in what would happen if the J and 8 of hearts were switched. Would the computer play double dummy?

[mrdct]

BunnyGo, on 2011-July-03, 16:35, said:

This hand was played in an automated free individual event. The clock ran out with 2.5 tricks to play, and was resolved in favor of declarer guessing correctly:

---

Note that if West pitches the diamond 8, then North has to guess if hearts are 1-1 or 2-0 at the end.

Now that I look at the hand more closely, I realize that there is no misguessing (as pitching the spade 4 and finessing, or ruffing the Ten and playing for 1-1 both work). However, I'm interested in what would happen if the J and 8 of hearts were switched. Would the computer play double dummy?

The robots never play double dummy (that would be cheating). My understanding is that the play methodology is to generate a largish number of hands to the current position which are consistent with the play thus far, work out the best double dummy card to play on each of them and then choose the card with the highest frequency.

[BunnyGo]

mrdct, on 2011-July-03, 23:04, said:

The robots never play double dummy (that would be cheating). My understanding is that the play methodology is to generate a largish number of hands to the current position which are consistent with the play thus far, work out the best double dummy card to play on each of them and then choose the card with the highest frequency.

Sorry, I was unclear. The problem is that time was called before the hand finished and a result was assigned. I was curious how this result is arrived at. Especially when there is still a guess for declarer to get right.

[barmar]

What I believe it does is set up the hands with GIB robots in all 4 seats, give them the bidding and play so far, and then let them play out the hands a bunch of times. They're playing single dummy, although they're using GIB's algorithm, which is based on double-dummy analysis of possible hands.

If there's a true 50/50 guess involved, then presumably the it will get it right about half the time. That "about" is critical: it's not likely to be exactly half (and if it plays an odd number of hands, it obviously can't be), so it's random which result will be more common. Although I've noticed in my playing against GIB that it has a tendency to stick in the missing queen (it uses double dummy analysis when defending, which assumes I'm going to guess right, which means it doesn't hurt for it to give it away). So I suspect in practice you'll be assigned the score that involves guessing correctly.

[Free]

BunnyGo, on 2011-July-03, 23:13, said:

Sorry, I was unclear. The problem is that time was called before the hand finished and a result was assigned. I was curious how this result is arrived at. Especially when there is still a guess for declarer to get right.

It's not really a guess, it's a percentage play here imo. If an automated score is assigned, AFAIK GIB determines a single dummy result.

I would suspect West will ruff singledummy because his partner may have the ♥T, in which case N-S can easily make the rest of the tricks.
If West doesn't ruff, then there's a guess and the generated deals to determine the best line of play will decide if North will let the ♣ win followed by en passant ruffing finesse or if he'll ruff and draw ♥K. Both lines win in 2 cases (letting ♣ win: West having Jx or J / ruffing the ♣: both 1-1 splits), so if there are more singleton ♥J's in East then GIB will ruff for example. But I expect it won't come to that anyway.