[Bbradley62]

I don't think I've seen GIB seem to forget that one of his cards is high. Advance to the 3-trick ending. East needs all 3 tricks to make his contract, which means he must find ♣Q. His ♥5 is good. If he's going to play for the Q to drop, he clearly should cash the heart first to ensure no worse than down 1. Why wouldn't he do that?

---

[johnu]

Probably due to the less than optimal way GIB does double dummy analysis. GIB assumes that double dummy it can play clubs for no losers, so playing the winning heart is irrelevant. Of course, look at how the ♥5 became high. Another case of GIB making a nullo play of sacrificing high cards for zero reason.

[1eyedjack]

johnu, on 2015-February-27, 19:21, said:

Another case of GIB making a nullo play of sacrificing high cards for zero reason.

On another day (not this hand, for sure) the certain knowledge that East started with 8 singleton or K8 doubleton when he plays the 8 on first round would concede more in the long term than its choosing to jettison a clearly irrelevant 8 this time whilst in possession of a lower card."... for zero reason"??

[johnu]

1eyedjack, on 2015-February-28, 03:33, said:

On another day (not this hand, for sure) the certain knowledge that East started with 8 singleton or K8 doubleton when he plays the 8 on first round would concede more in the long term than its choosing to jettison a clearly irrelevant 8 this time whilst in possession of a lower card."... for zero reason"??

Not sure what your point is if any adding your comment. South has already made a support double showing exactly 3 hearts, so East is known to have 4 exactly hearts.

If your point is that in 1 out of 100 (or whatever the percentage may be) hands, playing unnecessarily high cards may help the defense, nobody has denied that. If you think GIB has a plan when randomly tossing away high cards, maybe you can explain the reason behind tossing the ♥10 and ♥9 away and making East's spot card high.

[1eyedjack]

I think that you know perfectly well what my point is and are being deliberately obtuse, but I shall play along and spell it out (probably have before).

We might prefer the introduction of some signalling system, but let's face it, that ain't going to happen. Indeed the entire discussion could be futile if, as some suggest, BBO are not empowered to make changes.

The choices being considered here are

1) GIB's current policy: Play randomly from cards which it judges to be "small", that is to say, cards without (or with equal) trick-taking potential. The risk with this policy is that GIB may misjudge whether a card is "active" v "small".

2) Your preference: Play lowest from cards which are not trying to win a trick or force out a higher card from an opponent. The risk with this policy is that declarer can draw inferences from the particular spot card, particularly where that spot card is fairly high.

Policy (1) is preferable to policy (2) if the frequency of the risk with policy (2) materialising is greater than the corresponding frequency of the risk with policy (1) materialising.
Policy (2) is preferable to policy (1) if the frequency of the risk with policy (1) materialising is greater than the corresponding frequency of the risk with policy (2) materialising.

You suggest that the risk associated with policy (2) is about 1 in 100 hands, subject to a qualifier "or whatever the percentage may be". While accepting that there may in fact be some variation from 1 in 100, your opinion is clearly that it is somewhere in this ball-park. I think that you are out by an order of magnitude.

[johnu]

1eyedjack, on 2015-February-28, 14:00, said:

I think that you know perfectly well what my point is and are being deliberately obtuse, but I shall play along and spell it out (probably have before).

We might prefer the introduction of some signalling system, but let's face it, that ain't going to happen. Indeed the entire discussion could be futile if, as some suggest, BBO are not empowered to make changes.

The choices being considered here are

1) GIB's current policy: Play randomly from cards which it judges to be "small", that is to say, cards without (or with equal) trick-taking potential. The risk with this policy is that GIB may misjudge whether a card is "active" v "small".

2) Your preference: Play lowest from cards which are not trying to win a trick or force out a higher card from an opponent. The risk with this policy is that declarer can draw inferences from the particular spot card, particularly where that spot card is fairly high.

Policy (1) is preferable to policy (2) if the frequency of the risk with policy (2) materialising is greater than the corresponding frequency of the risk with policy (1) materialising.
Policy (2) is preferable to policy (1) if the frequency of the risk with policy (1) materialising is greater than the corresponding frequency of the risk with policy (2) materialising.

You suggest that the risk associated with policy (2) is about 1 in 100 hands, subject to a qualifier "or whatever the percentage may be". While accepting that there may in fact be some variation from 1 in 100, your opinion is clearly that it is somewhere in this ball-park. I think that you are out by an order of magnitude.

What you are apparently unable to understand is that the risks are not comparable.

By randomly sacrificing high cards you end up throwing away a trick like in this hand which is a clear loss. By playing the lowest card (and certainly you could randomize with something like 432 or 6432 and playing the 6 if you can reasonably analyze that playing the 6 won't cost a trick. If you can't rule out the 6 costing a trick, don't play it.) it's possible that opponents gain some information about your distribution but maybe it doesn't make any difference, and even if it could make a difference, maybe your distribution can be figured out from other bidding or play clues, or maybe playing for a particular distribution is the only way to maximize tricks. Gaining an advantage from knowing that GIB always plays the lowest spot card is usually going to take a parlay of low percentage circumstances that combine to be a very low percentage, including that declarer will play well enough to be able to formulate a viable plan of attack.

Given the state of GIB programming, analyzing whether randomly sacrificing a high card will or will not be costly is not within program parameters and apparently way to difficult to reprogram based on comments from BBO. On the other hand, I think always following suit with the lowest (or randomly from equals) is an attainable goal in the near future.

Edit: It's not just creating a winner out of the blue by throwing away high cards that's a problem. Throwing away high cards can give the opponents an extra entry that wasn't available, or enable a higher percentage line of play that wouldn't have been available if the location of the high card wasn't known. OK, maybe throwing high cards away leads to some kind of Grosvenor Coup, but in GIB's hands, it would clearly be accidental.

[1eyedjack]

johnu, on 2015-March-01, 13:28, said:

What you are apparently unable to understand is that the risks are not comparable.

Correct. I am unable to understand it precisely because the risks are indeed comparable, and nothing in the remainder of your post dissuades me from that view.

johnu, on 2015-March-01, 13:28, said:

By randomly sacrificing high cards you end up throwing away a trick like in this hand which is a clear loss.

Throwing the Heart 8 on this hand conceded precisely nothing to declarer. This hand is not a good example to illustrate either case. The only point of interest of this hand is that the defender failed to cash an established winner later in the play, which is a bug but an entirely different one.

johnu, on 2015-March-01, 13:28, said:

By playing the lowest card (and certainly you could randomize with something like 432 or 6432 and playing the 6 if you can reasonably analyze that playing the 6 won't cost a trick. If you can't rule out the 6 costing a trick, don't play it.) it's possible that opponents gain some information about your distribution but maybe it doesn't make any difference, and even if it could make a difference, maybe your distribution can be figured out from other bidding or play clues, or maybe playing for a particular distribution is the only way to maximize tricks. Gaining an advantage from knowing that GIB always plays the lowest spot card is usually going to take a parlay of low percentage circumstances that combine to be a very low percentage, including that declarer will play well enough to be able to formulate a viable plan of attack.

If you want me to accept that declarer will not always take a losing option when presented with a choice, you will get no argument from me. If you want to accept that there is no point in presenting declarer with a losing option because he might not take one, then you will.

The possibility that declarer might decline a losing option is certainly a factor to consider. My earlier post was simplified in the interests of brevity, and even then it was long enough. Do not fall into the trap of thinking that because I did not write a thesis I do not understand the subtleties beyond those expressed.

The more complete analysis would require multiplying the probability of each event with the financial consequences, which would include factorising the possibility that, presented with a losing option, declarer might not take it, and then integrating the result over the entire spectrum of possible outcomes.

johnu, on 2015-March-01, 13:28, said:

Given the state of GIB programming, analyzing whether randomly sacrificing a high card will or will not be costly is not within program parameters and apparently way to difficult to reprogram based on comments from BBO.

Don't make the mistake of assuming that comments on BBO are representative of a problem or of its scale. If a defender throws an active card and in the process costs a trick, you can expect a complaint to appear in the GIB forum. If a defender plays an 8 from 8xx and declarer erroneously plays him for a shortage, do not expect any posts in the forums extolling the virtues of how GIB did well.

[johnu]

1eyedjack, on 2015-March-01, 15:24, said:

Correct. I am unable to understand it precisely because the risks are indeed comparable, and nothing in the remainder of your post dissuades me from that view.

Wow, you do have your blinds on don't you. You don't see the difference between setting up a sure trick out of nothing, and hypothetically having the opponents fall for the falsecard, not having any other clues to the correct line of play based on the bidding and previous tricks, and then taking the wrong line of play.

1eyedjack, on 2015-March-01, 15:24, said:

Throwing the Heart 8 on this hand conceded precisely nothing to declarer. This hand is not a good example to illustrate either case. The only point of interest of this hand is that the defender failed to cash an established winner later in the play, which is a bug but an entirely different one.

What are you talking about? Declarer has ♥AQJ1097 in the combined hands. Of course playing the 8 doesn't cost anything. Would you say randomly playing the ♥10 and ♥9 didn't concede anything??? If not, please explain to me how the ♥5 ended up high at the end? About the only thing I agree with you in this entire thread is that not playing the ♥5 in the end game is a different type of bug.

1eyedjack, on 2015-March-01, 15:24, said:

Don't make the mistake of assuming that comments on BBO are representative of a problem or of its scale. If a defender throws an active card and in the process costs a trick, you can expect a complaint to appear in the GIB forum. If a defender plays an 8 from 8xx and declarer erroneously plays him for a shortage, do not expect any posts in the forums extolling the virtues of how GIB did well.

I've seen the problem first hand too many times to count. I don't need to read about it in these forums to know the frequency of the problem. As long as the human player knows that GIB can play random cards when following suit, it's not a matter of GIB doing well, it's just a matter of the human player guessing right or wrong which they may have done with or without a random card from GIB. If the human player does not know GIB will randomly play cards, then they will "misguess" more frequently.

[1eyedjack]

johnu, on 2015-March-01, 18:29, said:

Wow, you do have your blinds on don't you. You don't see the difference between setting up a sure trick out of nothing, and hypothetically having the opponents fall for the falsecard, not having any other clues to the correct line of play based on the bidding and previous tricks, and then taking the wrong line of play.

Ah, time to resort to the ad hominem attacks, I see. No I don't have my blinds on. Of course I see the difference. I take into account the frequency of an opponent failing to fall for a falsecard, and in my experience the remaining balance of occasions when he will fall for the falsecard (or to be more accurate, would benefit from GIB's known failure to falsecard in your alternative world) outstrips occasions when GIB throws a sure trick.

Where your blinds are firmly in place is that you will not accept that sometimes falsecards are effective, and that it is a valid question to determine how *that* frequency compares with trick chucks in concluding whether your change would be for the better.

You hold KT98 in trumps in dummy opposite A7654 in hand and you cash the Ace, all following, the Jack falling on your right. You continue with the 4 and LHO follows low. With nothing else to go on, do you finesse the second round or play for the drop? In the real world you finesse, and you would be right about twice as often as wrong. In your preferred GIB world the correct play is the drop, because GIB will always play J from QJ, and you would be right fractionally more than half the time. Is this how you want GIB to distort the game?

You hold AQ82 in dummy opposite K43 in hand, in NoTrumps and require 4 tricks in the suit. You cash the King, all following, RHO with the Jack. This being known to be RHO's lowest card you are now assured of 4 tricks by taking repeated deep finesses against LHO's T9, regardless of whether or not he splits them on the second round. The normal percentage play would be to play high when 2nd hand follows with a card lower than the 8 (absent other indicators, of course), and then resort to a finesse only on the third round of the suit if RHO follows with T or 9 on the second. Which may or may not be successful, but the success rate would fall somewhat below the 100% line of the second round deep finesse that GIB would grant us in your scenario

Both of these are mundane, bread-and-butter examples of issues that are presented to declarer day in day out. Do you see hands like this posted in the forums? Well, no, not least because GIB (thankfully) does not behave that way. Would you see these posts appearing if GIB were programmed to play bottom card always for fear of throwing an active card? Absolutely you would, and I submit that they would be rather more frequent than the posts that we currently get about chucking high cards. That is of course a guess on my part. Unless and until your proposal is implemented there will be no empirical data to support or refute it. It may be possible to predict it mathematically, but that would be beyond me.

johnu, on 2015-March-01, 18:29, said:

What are you talking about?

Yes you are right about this. The ♥10 did not cost but the ♥9 did. Playing the Ace at trick 1 didn't help the defence, and if South had unblocked the ♥J (the best double dummy defence which would never happen in your world if South had been a robot) then North I suspect would have got the later play right. Anyway I missed that so chalk this up as an example in your favour. Apologies for the distraction.

johnu, on 2015-March-01, 18:29, said:

I've seen the problem first hand too many times to count.

As have I, except for the "first hand" bit; only in the forums. I cannot remember the last occasion when GIB did it to me at the table. I am faced with restricted choice play problems pretty much every day (maybe nearly every tourney) where you are faced with them only 1 in 100 hands.

johnu, on 2015-March-01, 18:29, said:

If the human player does not know GIB will randomly play cards, then they will "misguess" more frequently.

I think it fair to assume that, whatever GIB's strategy is or will be in the future, at the time of play the human will(/should) know that strategy. If GIB will randomly play cards, it is unreasonable to consider the consequences of the human player not knowing that GIB will do so.

[johnu]

1eyedjack, on 2015-March-02, 00:47, said:

You hold KT98 in trumps in dummy opposite A7654 in hand and you cash the Ace, all following, the Jack falling on your right. You continue with the 4 and LHO follows low. With nothing else to go on, do you finesse the second round or play for the drop? In the real world you finesse, and you would be right about twice as often as wrong. In your preferred GIB world the correct play is the drop, because GIB will always play J from QJ, and you would be right fractionally more than half the time. Is this how you want GIB to distort the game?

Better reading comprehension by you would quickly resolve this disagreement problem. I mentioned playing randomly from equals when following suit is something that seems within the capabilities of the GIB programmers (but maybe not, just my opinion with having seen any of the code). Last time I checked, queen and jack were equals.

1eyedjack, on 2015-March-02, 00:47, said:

Yes you are right about this. The ♥10 did not cost but the ♥9 did. Playing the Ace at trick 1 didn't help the defence, and if South had unblocked the ♥J (the best double dummy defence which would never happen in your world if South had been a robot) then North I suspect would have got the later play right. Anyway I missed that so chalk this up as an example in your favour. Apologies for the distraction.

Playing the ♥A at trick 1 wins a trick which may not come back. South made a support double showing exactly 3 hearts and followed with the ♥J when the 3rd round of hearts was ruffed and North followed with ♥10. What would unblocking ♥J on the first heart ruff do?

1eyedjack, on 2015-March-02, 00:47, said:

I think it fair to assume that, whatever GIB's strategy is or will be in the future, at the time of play the human will(/should) know that strategy. If GIB will randomly play cards, it is unreasonable to consider the consequences of the human player not knowing that GIB will do so.

When I play in robot games, I almost always play in ACBL games where players pay US $1.00 per game. When I see unexpected results on a board, sometimes I'll check to see what happened at other tables. My estimate is that 1/4 - 1/3 the players are beginners/novices/players playing past their bedtime who don't seem to have any idea what their GIB opponents are doing, either in the bidding or playing. It's not advertised anywhere that GIB plays random cards, and even it was, most players probably wouldn't read the notice. Maybe only 10-20% are good enough to actually notice and keep track of the spot cards and consistently come up with a good percentage play when it matters.

[1eyedjack]

johnu, on 2015-March-02, 13:39, said:

Better reading comprehension by you would quickly resolve this disagreement problem.

Again with the ad hominem. Not to worry, I don't take office, but rather tend to gain confidence when you attack the messenger for want of an argument against the message. Be that as it may, I think that we are beginning to make progress.

johnu, on 2015-March-02, 13:39, said:

I mentioned playing randomly from equals when following suit is something that seems within the capabilities of the GIB programmers (but maybe not, just my opinion with having seen any of the code). Last time I checked, queen and jack were equals.

So you agree that it is acceptable for GIB to play at random from a choice of cards with equal trick-taking potential.

Now all that remains is for GIB to be programmed to recognise when two cards have equal trick taking potential, and not to jettison a high card that does not. I have no truck with that objective.

Deafening in its silence is your destruction of the second example

johnu, on 2015-March-02, 13:39, said:

When I play in robot games, I almost always play in ACBL games where players pay US $1.00 per game. When I see unexpected results on a board, sometimes I'll check to see what happened at other tables. My estimate is that 1/4 - 1/3 the players are beginners/novices/players playing past their bedtime who don't seem to have any idea what their GIB opponents are doing, either in the bidding or playing. It's not advertised anywhere that GIB plays random cards, and even it was, most players probably wouldn't read the notice. Maybe only 10-20% are good enough to actually notice and keep track of the spot cards and consistently come up with a good percentage play when it matters.

I'm sorry, but I really don't understand the point being made here. By all means apply some personal attacks for that. I am sure that I deserve it this time. My reading is that the argument for playing bottom card is that rabbits wouldn't notice one way or the other, right?

[Bbradley62]

And we've all seen this... GIB seems to assume that declarer knows where all the cards are and knows when to finesse vs not. Check out East's play on Trick 10 here:

---

[Antrax]

Bbradley62, on 2015-March-02, 14:17, said:

And we've all seen this... GIB seems to assume that declarer knows where all the cards are

Well yes, you know that's true. That's the shortcoming of double-dummy simulations, they assume all four players play double-dummy.

[johnu]

1eyedjack, on 2015-March-02, 14:12, said:

Again with the ad hominem. Not to worry, I don't take office, but rather tend to gain confidence when you attack the messenger for want of an argument against the message.

If it helps you, just consider it as constructive criticism and learn from it.