[pescetom]

Despite the repeated price hikes I have no illusions about robot improvements, but sometimes it seems they are even going backwards.
Maybe something about the double and possible trump distribution convinced it to blow the winning diamonds here?

[smerriman]

pescetom, on 2024-August-03, 14:43, said:

Maybe something about the double..

The double promises 3 cards in the unbid suits, including diamonds. Under that assumption, the robot's plan to throw two clubs seems perfectly fine to me.

[pescetom]

smerriman, on 2024-August-03, 17:35, said:

The double promises 3 cards in the unbid suits, including diamonds. Under that assumption, the robot's plan to throw two clubs seems perfectly fine to me.

Humans have been playing doubles of a minor oriented to the majors for some decades now... it doesn't seem unreasonable to seed the simulations with some doubt about the other minor.
But even assuming 3+ diamonds, I'm not so sure that it is advantageous to throw a second club loser under the inevitable ruff of a diamond winner... I would be curious to know (BBO having arbitrarily suppressed the Tricks function in Dealer) the result of a sufficiently numerous simulation.

[smerriman]

It would be nice for the robot to know humans play something else, but it always assumes they play GIB.

If you can provide an alternative line you think is better I can compare them in a simulation.. throwing the club seems good to me though; the club is always a loser, while a significant amount of time, it trades that for gaining a trump trick when the opponent ruffs. As always I could be missing something better so best to spell it out

[smerriman]

Actually, bridge aside, this was quite interesting from a mathematical point of view. Don't need to do a Monte Carlo simulation at all.

GIB's double is 12+ total points, 3-5 diamonds, 3-4 hearts, 3-4 spades, 2 or less clubs.

The J lead means West has precisely Kxxx.

4450 is ruled out, because despite the 3 points for the void, that's a max of 11 total points available.

Remembering that it docks a total point for a short suit containing an honor, that leaves only 8 technically distinct cases for West (though some turn out to be equivalent):

4342:
1) Kxxx-KQx-xxxx-?? (4+ hcp in clubs)
2) Kxxx-Kxx-xxxx-A? (6+ hcp in clubs)
3) Kxxx-Qxx-xxxx-AK (7 hcp in clubs)

4351:
4) Kxxx-KQx-xxxxx-? (3+ hcp in clubs)

4432:
5) Kxxx-KQxx-xxx-?? (4+ hcp in clubs)
6) Kxxx-Kxxx-xxx-A? (6+ hcp in clubs)
7) Kxxx-Qxxx-xxx-AK (7 hcp in clubs)

4441:
8) Kxxx-KQxx-xxxx-? (3+ hcp in clubs)

In situation 1, GIB's line takes 8 tricks, as does drawing a second round of trumps.

In situation 4 (least likely diamond split), drawing two rounds of trumps is needed to make 9 tricks, with GIB's line making 8 tricks.

However, in situations 2+3, GIB's line is required to make an overtrick.

In situation 5, GIB's line is the only line to make the contract.

In situations 6-8, GIB's line is required to make an overtrick (here you can get away with leading a second trump up, but then have to switch to GIB's line anyway).

So, independently from whether you like how it describes a double, given that, it looks like GIB's line was excellent here.

[pescetom]

smerriman, on 2024-August-05, 04:50, said:

Actually, bridge aside, this was quite interesting from a mathematical point of view. Don't need to do a Monte Carlo simulation at all.

GIB's double is 12+ total points, 3-5 diamonds, 3-4 hearts, 3-4 spades, 2 or less clubs.

The J lead means West has precisely Kxxx.

4450 is ruled out, because despite the 3 points for the void, that's a max of 11 total points available.

Remembering that it docks a total point for a short suit containing an honor, that leaves only 8 technically distinct cases for West (though some turn out to be equivalent):

4342:
1) Kxxx-KQx-xxxx-?? (4+ hcp in clubs)
2) Kxxx-Kxx-xxxx-A? (6+ hcp in clubs)
3) Kxxx-Qxx-xxxx-AK (7 hcp in clubs)

4351:
4) Kxxx-KQx-xxxxx-? (3+ hcp in clubs)

4432:
5) Kxxx-KQxx-xxx-?? (4+ hcp in clubs)
6) Kxxx-Kxxx-xxx-A? (6+ hcp in clubs)
7) Kxxx-Qxxx-xxx-AK (7 hcp in clubs)

4441:
8) Kxxx-KQxx-xxxx-? (3+ hcp in clubs)

In situation 1, GIB's line takes 8 tricks, as does drawing a second round of trumps.

In situation 4 (least likely diamond split), drawing two rounds of trumps is needed to make 9 tricks, with GIB's line making 8 tricks.

However, in situations 2+3, GIB's line is required to make an overtrick.

In situation 5, GIB's line is the only line to make the contract.

In situations 6-8, GIB's line is required to make an overtrick (here you can get away with leading a second trump up, but then have to switch to GIB's line anyway).

So, independently from whether you like how it describes a double, given that, it looks like GIB's line was excellent here.

Thanks for this and sorry for late reply.
Your analysis based upon 3+ diamonds assumption is impeccable, but unfortunately humans don't necessarily obey that and the robot result here was an absolute bottom.

[smerriman]

pescetom, on 2024-August-07, 12:38, said:

Your analysis based upon 3+ diamonds assumption is impeccable, but unfortunately humans don't necessarily obey that and the robot result here was an absolute bottom.

How many actually doubled as West? I would have doubled with 4423, though in that case both the trump + diamond lines go down (double dummy you need to work on spades), so the difference is just an undertrick, making the other line still better overall. Doubling with an 11 count, 4-3 in the majors and over half your points in the opponents' suit isn't on my radar at all, though humans do do very weird things. Unfortunately there's no forum for berating them

[pescetom]

smerriman, on 2024-August-07, 15:11, said:

How many actually doubled as West? I would have doubled with 4423, though in that case both the trump + diamond lines go down (double dummy you need to work on spades), so the difference is just an undertrick, making the other line still better overall. Doubling with an 11 count, 4-3 in the majors and over half your points in the opponents' suit isn't on my radar at all, though humans do do very weird things. Unfortunately there's no forum for berating them

I just checked... 3 out of 20 W (3 out of 18 humans) doubled. One of those NS made 2♥= on the same lead, another NS ended up in a suicidal 3NT after N redoubled and robot E bid diamonds.

I doubt many humans would fail to make 2♥ even if all W doubled.

[smerriman]

pescetom, on 2024-August-10, 04:21, said:

I doubt many humans would fail to make 2♥ even if all W doubled.

I put it to a poll at BW. So far 24 of the 33 votes are for playing the queen of spades at trick 1, which I hadn't paid enough attention to, thinking it was risky given you knew it was going to lose.. but perhaps a safer route, ignoring any chance of an overtrick and taking exactly 8 tricks. But of the 9 who voted to play the spade ace, 9/9 are taking GIB's line and going down on your hand.

[pescetom]

smerriman, on 2024-August-10, 20:52, said:

I put it to a poll at BW. So far 24 of the 33 votes are for playing the queen of spades at trick 1, which I hadn't paid enough attention to, thinking it was risky given you knew it was going to lose.. but perhaps a safer route, ignoring any chance of an overtrick and taking exactly 8 tricks. But of the 9 who voted to play the spade ace, 9/9 are taking GIB's line and going down on your hand.

Saw it yesterday, thanks. There were a few votes creeping in for the other options. Michael is way above BBO tournament level of course, but I think a few would arrive at playing the Queen. A whole load more would be playing hearts Ace and small, inferior though it is.

An important thing I overlooked and that is working against the robot is that most tables were playing the 2♥ contract in South not North, as most Italians will respond 1♦ whereas GiB responds 1♥, doubled or not. So they receive a clubs King lead and probably a club continuation even if they then find the switch to spades from East.

[TwoOneBEN]

To me it looks like declarer just went down for no reason. Playing trump at trick 9 can't be wrong as far as I can see.