[m1cha]

I just had this hand in the Free Daily (IMPs):

---

The final double is labeled "Takeout double -- 4-5 ♣; 4-5 ♦; 2-♥; 3-4♠; 18-20 total points". GIB partner passes with nothing but a J, and opps make 4♠X. If partner did what it was supposed to, which is bid its 5-card ♣, we play 5♣ which is cold. Does anyone believe I continue the tournament after my robot partner has just given me a ~ 20 IMP loss?

By the way, the takeout double does not show 3-4 ♠ when I promise 18+ points.
And at the bridge table I would have bid 4NT but here 4NT was something else; at least that's what the label said.

[1eyedjack]

Out of curiosity, how was 4N described? Not that I suggest that it is close to being right.

Much as I sympathise with your plight, I am surprised that it was a 20 IMP loss. Each action (bar North's final pass) is so routine that I would expect it to be close to flat. Perhaps not v par, but at least within the room.

[smerriman]

1eyedjack, on 2016-December-11, 14:40, said:

Much as I sympathise with your plight, I am surprised that it was a 20 IMP loss. Each action (bar North's final pass) is so routine that I would expect it to be close to flat. Perhaps not v par, but at least within the room.

I expect more of a 20IMP swing compared to the rare few who somehow got to 5♣. Though of course, if GIB hadn't passed the double, it wouldn't have been such a big swing.

GIB follows a very simple rule regarding doubles late in the auction that is easy to remember - a double is always for takeout, except for precisely the times you want it to be.

[1eyedjack]

smerriman, on 2016-December-11, 15:15, said:

I expect more of a 20IMP swing compared to the rare few who somehow got to 5♣. Though of course, if GIB hadn't passed the double, it wouldn't have been such a big swing.

GIB follows a very simple rule regarding doubles late in the auction that is easy to remember - a double is always for takeout, except for precisely the times you want it to be.

In order for there to be a swing, of whatever magnitude, it requires a deviation in the bidding or play.

For a given preceding sequence of bidding or play, and identical random number seed for simulations, GIB will always take the same action, however unreasonable you as a human might judge it. I should hope that within a given tournament the seed would be the same. That has not been questioned in the past.

The only individual at the table with the capability to create a diversion in the bidding or play is the sole human at the table, which is to say South. Once there has been such a deviation, it is of course open season, and that could be a deviation as trivial as playing a 2 where a 3 would have been an alternative. There are fewer opportunities for such trivial deviations in the bidding, the first of which must, as mentioned, come from the human South.

Your expectation of "more of a 20IMP swing" requires that you judge one or more of South's actions to be abnormal. Which?

[smerriman]

1eyedjack, on 2016-December-11, 15:32, said:

Your expectation of "more of a 20IMP swing" requires that you judge one or more of South's actions to be abnormal.

Nope - just one player needs to have abnormal actions, and if you've played any of the daylong tournaments, you'll know there's always variation no matter how simple the hand looks.

For the sake of argument, if a single player made an abnormal bid somewhere and gets to 5♣ making, and the rest of the field are in 4♠x, that's an apparent 15 IMP swing.

Though as mentioned, GIBs involvement in that swing is only a facade, as if GIB hadn't passed the double, nobody would be in 4♠x.

[1eyedjack]

smerriman, on 2016-December-11, 16:17, said:

Nope - just one player needs to have abnormal actions,

There is no inconsistency here. The point is that that "one player" needs to be South, the only human at the table. All the others, being robots, are incapable of taking "abnormal actions"; or at least they will all without fail take the same abnormal action for the given input data.

So my point stands.

[smerriman]

You said it's not possible for a 20IMP swing without one of the above bids by South being abnormal.

I'm saying even if 99% of players make the same bids as in the above auction, it's still possible (and likely) for 5♣ to outscore 4♠x by a huge margin, such as the 20IMP which m1cha referred to.

Someone may have misclicked; someone may have overcalled instead of doubled, and so on. These are all bad and abnormal actions, but out of 100 random BBOers, they always occur.

[iandayre]

1eyedjack, on 2016-December-11, 16:27, said:

There is no inconsistency here. The point is that that "one player" needs to be South, the only human at the table. All the others, being robots, are incapable of taking "abnormal actions"; or at least they will all without fail take the same abnormal action for the given input data.

So my point stands.

I doubt if very many Souths passed 3S. 4C certainly isn't a strange choice.

[iandayre]

Deleted Duplicate

[m1cha]

1eyedjack, on 2016-December-11, 14:40, said:

Out of curiosity, how was 4N described?

Sorry but I forgot it the moment I saw it was useless. If I had to guess now I would say balanced 25+ tp but it's really not more than a guess.

1eyedjack, on 2016-December-11, 14:40, said:

Much as I sympathise with your plight, I am surprised that it was a 20 IMP loss. Each action (bar North's final pass) is so routine that I would expect it to be close to flat. Perhaps not v par, but at least within the room.

I can't say what loss it actually was or will be. If I had finished the tournament, I wouldn't get the result before tomorrow. What I estimated was -590 on the one side and +600 on the other side which I had got if GIB had passed as in should have. With a par in the middle that would be -11 IMPs vs. +12 IMPs, a difference of 23. With par at -420 points it would be -5 IMPs vs. +15 IMPs, a difference of 19. Of course you are right that if all players have this problem, the difference goes down to zero. But that would not be good news because the smaller the difference the more players were upset by GIB.

1eyedjack, on 2016-December-11, 15:32, said:

In order for there to be a swing, of whatever magnitude, it requires a deviation in the bidding or play.

For a given preceding sequence of bidding or play, and identical random number seed for simulations, GIB will always take the same action, however unreasonable you as a human might judge it. I should hope that within a given tournament the seed would be the same. That has not been questioned in the past.

I guess it's correct. When I have noticed deviations in play so far, they were usually caused by one player playing the 3 and another player playing the 5. I remember a forum discussion in which case nondeterministic behavior was observed because boards used for comparison were played by older versions if GIB.

If GIB were an AI, it would have to vary the rules in order to learn. But we don't seem so see this.

1eyedjack, on 2016-December-11, 15:32, said:

Your expectation of "more of a 20IMP swing" requires that you judge one or more of South's actions to be abnormal. Which?

In a free weekly I'd expect a two-digit percentage of players to bid 1♦ instead of double, and a few less to bid 2NT. The final double also seems far from clear to me in this vulnerability. Perhaps I'll continue to play just to see how this one ends up.

smerriman, on 2016-December-11, 16:17, said:

Nope - just one player needs to have abnormal actions, and if you've played any of the daylong tournaments, you'll know there's always variation no matter how simple the hand looks.

For the sake of argument, if a single player made an abnormal bid somewhere and gets to 5♣ making, and the rest of the field are in 4♠x, that's an apparent 15 IMP swing.

Correct me if I'm wrong but if 1 player reaches +600 and 99 others play in -590, then this one player gets +15 IMPs and those 99 get -0.15 IMPs, don't they? This scenario wouldn't bother me but from my experience with the Free Dailys I'd guess the percentage of players in 4♠X is less than 40 % and then, in addition, there's the question of how many of these will bring the contract down.

1eyedjack, on 2016-December-11, 16:27, said:

There is no inconsistency here. The point is that that "one player" needs to be South, the only human at the table. All the others, being robots, are incapable of taking "abnormal actions"; or at least they will all without fail take the same abnormal action for the given input data.

Yes but this is not a national championship, it's a Free Daily, and many human players there love "abnormal actions".
In some cases EW may even play for a part score.
Edit: I finished the tournament and will let you know the result of this board tomorrow (though late).

[1eyedjack]

smerriman, on 2016-December-11, 16:31, said:

You said it's not possible for a 20IMP swing without one of the above bids by South being abnormal.

I'm saying even if 99% of players make the same bids as in the above auction, it's still possible (and likely) for 5♣ to outscore 4♠x by a huge margin, such as the 20IMP which m1cha referred to.

Someone may have misclicked; someone may have overcalled instead of doubled, and so on. These are all bad and abnormal actions, but out of 100 random BBOers, they always occur.

I was referring to your average IMP score over the room. Sure, you never get a completely flat board, especially in a room that size. If you choose to exclude most of the population in favour of comparing your result with the fringe, you would be right, but why would you do that?

Of course, it is a missed opportunity, even if the hand was completely flat.

[1eyedjack]

iandayre, on 2016-December-11, 16:35, said:

I doubt if very many Souths passed 3S. 4C certainly isn't a strange choice.

Wow. I suppose you are right about passing 3S. But 4C??

[smerriman]

1eyedjack, on 2016-December-11, 23:46, said:

Wow. I suppose you are right about passing 3S. But 4C??

Here's the traveller if you want to see the wide range of bidding that went on.. http://webutil.bridg...49948-R1-B2-I28

(Incidentally, the difference between 5♣= and 4♠x= was 20.67 IMP - nice estimate )

[Stephen Tu]

To me I think you are being result merchant. Do we really want North to pull with just 5 card clubs and 1 jack? What if East has much more common 4621 type shape rather than 5611, and put ck offside instead on. Now both contract fail. How often this happen, and how often 5c goes for 500 vs. nothing, vs. this double game swing?

For me if you pass 3s and double 4s you are saying you think can beat it (takeout double of heart, penaltyish of spade), if my partner does this I am never going to pull this, from my point of view partner has strong hand and spade break 4-1.

To me just unlucky hand. If South unilateral 4nt over 4s for minor, or North pull south "takeout double", yes you win this hand but I think long run not going to work so well.

Also, there is the matter that you should beat 4s instead of letting it make. All you have to do is tap declarer at some point instead of leading spade after cashing HA, he can't ruff hearts out anymore.

[m1cha]

Stephen Tu, on 2016-December-12, 04:33, said:

To me I think you are being result merchant. Do we really want North to pull with just 5 card clubs and 1 jack?

This is not about pulling a penalty double. This is about converting a takeout double into a penalty double, and for this you need good trumps and points. North has neither.

All I want is that if the labels say this is a takeout double, GIB treats it like a takeout double.

Stephen Tu, on 2016-December-12, 04:33, said:

Also, there is the matter that you should beat 4s instead of letting it make. All you have to do is tap declarer at some point instead of leading spade after cashing HA, he can't ruff hearts out anymore.

Yes, perhaps I should have beaten it just as half of the field did, and it looks easy to beat once I have seen the full board. But part of my point is that after what happend I get the cards wrong and tend to play N for less clubs and more hearts. If the minor suit honors are to my left, playing the minors now does not look like a good idea to me. While pullilng trumps is good if partner has less clubs and consequently more hearts.

[Stephen Tu]

m1cha, on 2016-December-12, 09:20, said:

[/size]
This is not about pulling a penalty double. This is about converting a takeout double into a penalty double, and for this you need good trumps and points. North has neither.

All I want is that if the labels say this is a takeout double, GIB treats it like a takeout double.

You need good trumps to convert *low level* contracts when partner has asked you to "normally take this out". It is absolutely not true that you need points and trumps to convert *high level* doubles when partner has shown considerable strength himself and usually enough to set the contract himself and hasn't made a takeout double of the second suit, but just shown extra values with *length* in the trump suit, and in my mind has told you *I think I got this set myself*!

Pull every time just because weak has to be weighed for probability they go set, vs. probability you make. It's usually a lot easier to take 4 tricks than 11 when you only have half the deck in high cards.

Your 2nd double is still takeout *of hearts*. You don't have a takeout double *of spades*. The description of the 2nd double is still showing heart shortness and spade length. The second double is thus penalty of spades. It's not supposed to be I want you to bid a minor. There's really no hand that can really justify that, that shouldn't have taken a different action on earlier rounds.

Arguably the "takeout double" description should be removed, as it is a leftover description from earlier and isn't takeout of the current contract, but I agree with the behavior of the North robot in response. But this is just a wording description bug as the suit length and point range of the South double are reasonable. It didn't suddenly become spade shortness!

[m1cha]

Stephen Tu, on 2016-December-12, 11:21, said:

Pull every time just because weak has to be weighed for probability they go set, vs. probability you make. It's usually a lot easier to take 4 tricks than 11 when you only have half the deck in high cards.

That depends very much on the distribution. And if I act over opponents' 4♠ while my partner has passed all the time, the distribution can be wild.

Stephen Tu, on 2016-December-12, 11:21, said:

Your 2nd double is still takeout *of hearts*. You don't have a takeout double *of spades*. The description of the 2nd double is still showing heart shortness and spade length. The second double is thus penalty of spades. It's not supposed to be I want you to bid a minor. There's really no hand that can really justify that, that shouldn't have taken a different action on earlier rounds.

No. The second double shows 18+ points according to the label. And as we all know very well, once I have shown 18+ points, my first double is NOT a take-out double of hearts any more but just a strength-showing bid. Consequently if the second double is still a take-out double, it is a take-out double of spades. If it is not a take-out double of spades, it is not a take-out double at all. I admit however that the length description in the label suggests otherwise.

Stephen Tu, on 2016-December-12, 11:21, said:

Arguably the "takeout double" description should be removed, as it is a leftover description from earlier and isn't takeout of the current contract, but I agree with the behavior of the North robot in response. But this is just a wording description bug as the suit length and point range of the South double are reasonable. It didn't suddenly become spade shortness!

If it is a take-out double at 18+ points, it did suddenly become a spade shortness, and there is nothing strange about it. But I do agree with the rest of your paragraph and I think this is where we will end up. I don't argue against a penalty double over 4♠ and if it is a penalty double, North's action is certainly correct. But if that's the case, it should be called penalty, the "takeout double" description is misleading and should please be removed - as always when I show 18+ points after a take-out double.

One further wish if the double is for penalty: 4NT should please show a 2-suiter. (I'm afraid this will be more complicated than changing the label but I feel it would be welcome by 90 % of the players.)

[Stephen Tu]

m1cha, on 2016-December-12, 17:58, said:

That depends very much on the distribution. And if I act over opponents' 4♠ while my partner has passed all the time, the distribution can be wild

Not when you passed over a non-forcing 3 spades!!! To me if you pass over 3S showing you are OK with defending 3s so when you double 4S your bid is like 100% for penalties. How can you have hand willing to defend 3 but not 4? Seriously take the North hand and auction and poll somewhere after 4s-x, the number of people who actually take it out should be miniscule and IMO don't understand bidding. You don't have a monster minor hand and pass 3 spades or start with a takeout double. Even with 18 hcp. If you had a huge hand both minors you would start with 2nt, or maybe 4nt. With a huge hand and one minor you would bid 4m or 5m over 3 spades, or have started with like 3nt overcall or 1H-3H jump cue looking for a stopper.
.

Quote

No. The second double shows 18+ points according to the label. And as we all know very well, once I have shown 18+ points, my first is NOT a take-out double of hearts any more but just a strength-showing bid. Consequently if the second double is still a take-out double, it is a take-out double of spades. If it is not a take-out double of spades, it is not a take-out double at all. I admit however that the length description in the label suggests otherwise.

No it's not a takeout double of spades at all, as I said that is leftover from earlier round. It was takeout of hearts, now it is basically penalty. You really can not just have arbitrary shape at this point, as minor hands would bid differently on earlier round. I suppose you can have a good 18+/19+ balanced hand that was too strong to overcall 1nt, but a the 4s-x point now are doubling based on defensive tricks expecting partner to pass 100%.

Quote

If it is a take-out double at 18+ points, it did suddenly become a spade shortness, and there is nothing strange about it. But I do agree with the rest of your paragraph and I think this is where we will end up. I don't argue against a penalty double over 4♠ and if it is a penalty double, North's action is certainly correct. But if that's the case, it should be called penalty, the "takeout double" description is misleading and should please be removed - as always when I show 18+ points after a take-out double.

Yes the word description as I said is incorrect leftover, common thing in how GIB generate description, but the shape/strength description should have given you a clue of how North should act.

Quote

One further wish if the double is for penalty: 4NT should please show a 2-suiter. (I'm afraid this will be more complicated than changing the label but I feel it would be welcome by 90 % of the players.)

4nt IMO shouldn't exist at this point, I suppose you can define it this way if you want. With both minors should have done other action earlier. 4nt = try to commit suicide, hope for miracle layout like you found. You seriously think your expected value with South hand is greater by sacrificing over 4s rather than sawing it off, or even defending undoubled, considering the universe of all hands E/W can hold on the auction, not just the one actual deal?

[m1cha]

Stephen Tu, on 2016-December-12, 21:39, said:

Not when you passed over a non-forcing 3 spades!!! To me if you pass over 3S showing you are OK with defending 3s so when you double 4S your bid is like 100% for penalties. How can you have hand willing to defend 3 but not 4?

Because 4M counts more than 3M? To me pass over 3♠ means nothing but that I'm okay if they play that. It can mean, among other things, that I don't want them to bid 4♠ because I think they'll make it. And for this reason I may defend over 4♠ while I don't bid over 3♠. Not at red vs. white, I admit, but the bidding rules of GIB don't consider this.

Stephen Tu, on 2016-December-12, 21:39, said:

Seriously take the North hand and auction and poll somewhere after 4s-x, the number of people who actually take it out should be miniscule

I believe you. At the bridge table. Also at GIB when the label says it's a penalty, of course, but not in GIB when the label says it's for takeout.

Stephen Tu, on 2016-December-12, 21:39, said:

Yes the word description as I said is incorrect leftover, common thing in how GIB generate description, but the shape/strength description should have given you a clue of how North should act.

Yes I might have found the clue and interpreted the wrong label correctly. But certainly this is not how you want GIB to work.

This is why I suggest to remove that leftover, and I have no other complaint about that board.

[WesleyC]

I'm not qualified to comment on the GIB labels (which is obviously incorrect in many cases) but if you think a bit more about it, bridge logic should lead you to the conclusion that your double of 4S should be for penalties.

Basically the rule is, if you've had a chance to make a takeout double of a suit but chose not to then later double that suit at a higher level, then double is penalties.

Holding significant extra values, great minors and hardly any values in Spades, you should make a takeout double of 3S which would allow your partner to take the (making) sacrifice in 5C.

Passing 3S and then doubling 4S should be consistent with a hand like AQJT x Axxx Kxxx where the opponents have walked into a bad break and are about to be punished!