[pescetom]

It did it again at first chance

Bottom of 24 tables, all but 1 in 3NT.

Just how many diamonds distributions did it simulate?

[pescetom]

Bump

(if one posts this kind of hand on that other forum, at least there is the illusion of a promise that the resident AI expert will look into it)

[smerriman]

If I feed the diamond suit into SuitPlay, it tells me the best MP line is precisely the one GIB took - deep finesse, then after West wins the J, play for the drop.

So once again, I ask you to outline what you believe the problem is first, unrelated to what other tables did..

[pescetom]

 smerriman, on 2025-March-27, 18:21, said:

If I feed the diamond suit into SuitPlay, it tells me the best MP line is precisely the one GIB took - deep finesse, then after West wins the J, play for the drop.

So once again, I ask you to outline what you believe the problem is first, unrelated to what other tables did..

Are you sure that is the best line to take 4 tricks in the suit, not 5? Opponents are not going to take huge risks for an overtrick, even at MP.

[smerriman]

No, it says it is the best line at MPs. It gives different lines for the best chance of 5 tricks (low to queen), and the best chance of 4 tricks (ace then low to queen). But the original line outperforms both at MPs.

And you didn't answer the question I don't ask to imply GIB is right, but to find out what you believe is wrong. You wouldn't believe how many hours I would save at work if clients could just start by providing accurate information of what they think isn't working and what should be happening, instead of "I think something is wrong, can you check?"

[pescetom]

 smerriman, on 2025-March-29, 04:38, said:

No, it says it is the best line at MPs. It gives different lines for the best chance of 5 tricks (low to queen), and the best chance of 4 tricks (ace then low to queen). But the original line outperforms both at MPs.

And you didn't answer the question I don't ask to imply GIB is right, but to find out what you believe is wrong. You wouldn't believe how many hours I would save at work if clients could just start by providing accurate information of what they think isn't working and what should be happening, instead of "I think something is wrong, can you check?"

Ok, but I don't think the choice of best line at MPs can be independent of the situation in other suits and the preceding tricks.

No other player in the tournament chose the same line as GiB and I see no good reason why one should in this situation. Are you really arguing that a world class player would do so against these players or against his peers?

I did implicitly answer your question: I think it is wrong to unnecessarily risk losing control twice in the suit, given the situation in other suits. If you want to know what the alternative is, I can only say that at IMPs I would have played A and then low to Q, here at MP I would have started low to Q and then decided how to continue: probably laying down A, but I think most of the players in this tournament would return to hand and finesse.

[smerriman]

 pescetom, on 2025-March-29, 07:38, said:

Are you really arguing that a world class player would do so against these players or against his peers?

I specifically said the exact opposite; that I am not saying GIB made a world class play, but that I wanted you to explicitly state what you believe a world class player would do so that I could compare that exact line to GIB and investigate whether you were correct or not, and/or why GIB might not take that better line. I had no idea what you were suggesting implicitly, since I am not a world class player.

Now that you finally have, I can look

[pescetom]

smerriman, on 2025-March-29, 12:34, said:

I specifically said the exact opposite; that I am not saying GIB made a world class play, but that I wanted you to explicitly state what you believe a world class player would do so that I could compare that exact line to GIB and investigate whether you were correct or not, and/or why GIB might not take that better line. I had no idea what you were suggesting implicitly, since I am not a world class player.

Now that you finally have, I can look

Thanks, but I don't think I said or even suggested what a world class player would do, I honesty don't know
I said only what I would not do, the other players in this tournament would not do, and (unless I am missing something) what a world class player would not do, namely what GiB did and you say Suitplay endorses.
We mortals would start with small to Q, if only out of mental laziness.

[smerriman]

Assume for now that all tables lead a spade and switch to a club. (You will disagree, but disagree later.)

If other tables would do what you think they will, playing low to the queen, then finessing on the second round if West plays an honor:

- if East holds Kxx as actually happened, they take 10 tricks and GIB takes 7 - disaster, must be a bug!
- but if East holds xx (marginally more likely), we get the exact opposite, with GIB taking 10 tricks and others taking 7 - oops, disaster for all humans.
- if East holds Kx, GIB loses a trick
- but if East holds KJx or KJxx, GIB gains a trick

Low to the queen then the ace gives a similar story, except it's Jxx where you're getting 7 tricks.

So unless I've missed a case somewhere, both of these seem clearly inferior.

Of course, all of these lines are so close that some luck was involved that GIB stumbled into the SuitPlay-optimal line.

I'm guessing (again) that your argument is therefore not related to any of this, but based on the fact not all other tables would have led a spade? That could well be a valid argument, though not something GIB takes into account; it assumes other tables have the same choice it does.

But if not for that, this type of scenario is exactly what I was meaning earlier where a score of 0% may be more due to your suggestion of "mental laziness" by all humans, rather than a guaranteed GIB bug.

[pescetom]

smerriman, on 2025-March-29, 14:20, said:

Assume for now that all tables lead a spade and switch to a club. (You will disagree, but disagree later.)

If other tables would do what you think they will, playing low to the queen, then finessing on the second round if West plays an honor:

- if East holds Kxx as actually happened, they take 10 tricks and GIB takes 7 - disaster, must be a bug!
- but if East holds xx (marginally more likely), we get the exact opposite, with GIB taking 10 tricks and others taking 7 - oops, disaster for all humans.
- if East holds Kx, GIB loses a trick
- but if East holds KJx or KJxx, GIB gains a trick

Low to the queen then the ace gives a similar story, except it's Jxx where you're getting 7 tricks.

So unless I've missed a case somewhere, both of these seem clearly inferior.

Of course, all of these lines are so close that some luck was involved that GIB stumbled into the SuitPlay-optimal line.

I'm guessing (again) that your argument is therefore not related to any of this, but based on the fact not all other tables would have led a spade? That could well be a valid argument, though not something GIB takes into account; it assumes other tables have the same choice it does.

But if not for that, this type of scenario is exactly what I was meaning earlier where a score of 0% may be more due to your suggestion of "mental laziness" by all humans, rather than a guaranteed GIB bug.

I'm really not so much interested in establishing yet another guaranteed GiB bug, more worrying that it delivered three consecutive unexpected bottoms in one of the situations where it would usually do better than a pickup human partner. In a moment when BBO is finally changing GiB, this is particularly disconcerting and I would like to be reassured it is not a regression (or even a Basic Robot).

But yes, I do think most humans would lead spades (even if someone somehow found clubs), and that GiB should take it into account.

I haven't thought through how close the various lines are, although I do suspect that the line chosen is distinctly inferior to starting with low to Q in the circumstances.

But even just looking at how the situation evolved after that choice, on trick 6 when 7♦ appears in East, does it make sense to put up the A rather than Q?
There is only the K missing: if W can cover then we still have 6 more tricks plus the 3 already taken. Unless I am missing something, this is risking a quite probable -2 in the hope of +1, when making is certain.

[johnu]

pescetom, on 2025-March-28, 09:38, said:

Are you sure that is the best line to take 4 tricks in the suit, not 5? Opponents are not going to take huge risks for an overtrick, even at MP.

The best line for 4 tricks is to play ♦A at trick 1, making 5 tricks when ♦K is singleton is offside, making 4 tricks whenever KJx(x) is not offside where the maximum number of tricks is 3.

pescetom, on 2025-March-29, 15:44, said:

I haven't thought through how close the various lines are, although I do suspect that the line chosen is distinctly inferior to starting with low to Q in the circumstances.

But even just looking at how the situation evolved after that choice, on trick 6 when 7♦ appears in East, does it make sense to put up the A rather than Q?

Answering the 2nd question first, yes, playing for diamonds to split is better than finessing a 2nd time (basically, the law of vacant spaces, e.g. chance of 2-0 split is 48%, chance of 1-1 split is 52%).

As to the 1st question, as noted, the initial double finesse is SuitPlay's MP line. I looked at the numbers and don't understand why that is the best MP line. Playing to the ♦Q actually makes 5 tricks a couple percent more than the initial finesse of ♦10, and both produce 3 tricks at the same rate.

[smerriman]

pescetom, on 2025-March-29, 15:44, said:

I haven't thought through how close the various lines are, although I do suspect that the line chosen is distinctly inferior to starting with low to Q in the circumstances.

I'd like to know why, so please post once you've thought about it.

johnu, on 2025-March-29, 17:09, said:

I looked at the numbers and don't understand why that is the best MP line. Playing to the ♦Q actually makes 5 tricks a couple percent more than the initial finesse of ♦10, and both produce 3 tricks at the same rate.

The cases you take 3 tricks cancel out, but Kx onside isn't as common as KJx(x) onside.

[johnu]

smerriman, on 2025-March-29, 19:00, said:

The cases you take 3 tricks cancel out, but Kx onside isn't as common as KJx(x) onside.

Yes, but finessing the queen also wins in the actual case, Kxx with singleton J offside.

Double finesse makes 5 tricks
x and KJx 12.43%
void and KJxx 4.78%
xx and KJ (not a double finesse, but playing the queen on the jack or ace on the king) 6.78%

Total 23.99%

Finessing the Queen at trick 1
Jx and Kx - 13.57%
J and Kxx - 6.22%
xx and KJ - 6.78%

Total 26.57%

[smerriman]

You're missing East holding Jxx where finessing the queen loses (if you're planning to follow up by playing for the drop. If you were going to finesse twice, it's xx with East which loses instead).

[pescetom]

 johnu, on 2025-March-29, 17:09, said:

Answering the 2nd question first, yes, playing for diamonds to split is better than finessing a 2nd time (basically, the law of vacant spaces, e.g. chance of 2-0 split is 48%, chance of 1-1 split is 52%).

Thanks, but is that enough to make playing for the 1-1 split better, in the circumstances?
See what I wrote immediately afterwards, or let me now rephrase my question in your terms: is it better to play for 3NT+1 52% 3NT-2 48% or for 3NT+1 48% 3NT= 52%?

[smerriman]

pescetom, on 2025-March-30, 09:17, said:

is it better to play for 3NT+1 52% 3NT-2 48% or for 3NT+1 48% 3NT= 52%?

Where do you get these numbers from? If you play the Q on the second round and it loses, the opponents cash three further clubs and you still get 3NT-2.

But in general, a 52% chance of 100% and a 48% chance of 0% is surely better than a safe 50%, unless you know you're winning the tournament by so much that you don't want to risk a 0 (though in reality you may still be risking if it the other top players go for the best line). But tournament-level logic is even more beyond GIB than what-other-tables-might-do logic.