[carmelbobc]

GIB passes obvious TOD after RHO opens a weak 2H, which responder raises to 3H. Here is the board:

https://tinyurl.com/y5k3xn24

To add salt to the wound, GIB then makes the only lead that allows the opponents to make!

Very much will appreciate someone telling me why. Certainly seems like a bug to me!

[Stephen Tu]

While GIB has problems with defining and reacting to doubles in a ton of auctions, I don't think this particular deal is necessarily one of them. The definition of the double is reasonably correct; support for other suits, 2- H, 14+ TP. Maybe it should be defined as slightly stronger but it's more or less OK.

Now, why does it pass? Generally, the way GIB operates is it deals a bunch of hands, filtering for ones that match the given auction. Then it chooses the bid/play that works out best double dummy opposite the sum total of hands. Bridge being a probabilistic game, sometimes the action that works out best on average can work out very badly on any one hand.


Is pass justifiable? Maybe. It has 2 bullets, no spade fit, and not one especially long minor. It expects you to have 2 trumps usually, since East only bid 3. It's not unreasonable to calculate that 5m is fairly unlikely to make when holding only 9 count + 2 heart losers off top most of the time. Nor that 4c might also go down with some frequency. It might well be easier to take 5-6 tricks in H. It's expecting to score spade tricks a lot more often; it doesn't expect the spade split to be 7-0! Often it is expecting your side to take 3 spade tricks or 2 spade tricks and a ruff, rather than zero. So maybe it is passing expecting a lot of +100s, some +300s, while it thinks 4c is going down sometimes, and successful 5c is rarely going to be reached, and bidding 5c by itself goes down too often also to be profitable. I personally think GIB might well be right here to pass on average.

On the lead, well again it's a probabilistic decision, and I think a spade lead is absolutely normal, though not best on this deal. Again it didn't think spades were going to be 7-0.

Imagine how you would do if east were some more pedestrian 4342 type shape; pass might look a lot more reasonable!


BTW, although it found the wrong lead on this deal, GIB declarer didn't find the double dummy line to make, and gave you another chance. You could have set it by just forcing a ruff in dummy after winning the 2nd club, instead of shifting to the trump A, which would have prevented declarer from scoring multiple spades.

[carmelbobc]

Thank you so very much, Stephen. I learned a lot from your post; it really makes sense. And, I have a renewed respect for my BOT partners! On that board, had I followed the old adage to return partner's opening lead, we would have set them and scored very well.

I suggest BBO include your description of how the BOTs decide whether a double is takeout or not in its materials.

[Stephen Tu]

 carmelbobc, on 2019-July-15, 13:22, said:

I suggest BBO include your description of how the BOTs decide whether a double is takeout or not in its materials.

My description above really isn't how the bot decides whether a double is takeout or not. That's in other posts like my response to rustysnow's recent post. It decides whether a double is takeout or not based on its bidding rules database, a collection of thousands of programmer written rules that tell it how to interpret each auction. Given the impossibly huge # of possible competitive auctions, and today's complex norms of which doubles should be takeout vs penalty, these rules only cover a small percentage of auctions, and have massive coverage gaps and errors. So especially on complicated rare auctions, double will often have some not very well defined meaning, unclear whether takeout or penalty, and one is at a guess what to do if robot doubles, and also at a guess whether the robot will leave your double in when you want it to, or take it out when you want it to. It's a hard problem that BBO hasn't made much progress on, the robots are just awful compared to good humans in interpreting doubles.

My first post in this thread is more "given that the database correctly defined the double as takeout, why it decided to leave it in for penalties anyway without a trump stack, which is more often right as the auction creeps higher, than it is at the one or two levels".

Quote

On that board, had I followed the old adage to return partner's opening lead, we would have set them and scored very well.

To get better, you have to try defend by calculation (picturing declarer's hand, counting out the board, playing out the rest of the tricks in your head), not by following adages. On this board, you should know that declarer has 6 trumps (from the opening), and that he will be able to ruff the spades good (spade position has already been revealed) and get there on the third round of trumps if you just play trumps. So the only hope is to shorten dummy's trumps and keep the trump ace to avoid declarer establishing spades and drawing trumps ending in dummy.

Following adages should be mostly confined to when you haven't gathered enough information about the hand yet such that there are still multiple options and you have to guess what to do. Then you follow adages because they tend to reflect percentage actions when there are still lots of unknowns.

I highly recommend the books "How to Defend a Bridge Hand", by Bill Root, and later "Killing Defence at Bridge" by Hugh Kelsey if you wish to improve your game.
Eddie Kantar also has some good defense books, and good quiz books "Kantar for the Defense".

[carmelbobc]

Excellent advice. Thank you.

Question: Might there be a way in which GIB can perform that simulation as soon as S views how a double will be interpreted? Or, give S an option to indicate that he wants the double to be for penalty?

By the way, when I click to "Reply," I do not see a panel in which to do so. So, I'm just adding my reply here.

 Stephen Tu, on 2019-July-15, 15:31, said:

My description above really isn't how the bot decides whether a double is takeout or not. That's in other posts like my response to rustysnow's recent post. It decides whether a double is takeout or not based on its bidding rules database, a collection of thousands of programmer written rules that tell it how to interpret each auction. Given the impossibly huge # of possible competitive auctions, and today's complex norms of which doubles should be takeout vs penalty, these rules only cover a small percentage of auctions, and have massive coverage gaps and errors. So especially on complicated rare auctions, double will often have some not very well defined meaning, unclear whether takeout or penalty, and one is at a guess what to do if robot doubles, and also at a guess whether the robot will leave your double in when you want it to, or take it out when you want it to. It's a hard problem that BBO hasn't made much progress on, the robots are just awful compared to good humans in interpreting doubles.

My first post in this thread is more "given that the database correctly defined the double as takeout, why it decided to leave it in for penalties anyway without a trump stack, which is more often right as the auction creeps higher, than it is at the one or two levels".



To get better, you have to try defend by calculation (picturing declarer's hand, counting out the board, playing out the rest of the tricks in your head), not by following adages. On this board, you should know that declarer has 6 trumps (from the opening), and that he will be able to ruff the spades good (spade position has already been revealed) and get there on the third round of trumps if you just play trumps. So the only hope is to shorten dummy's trumps and keep the trump ace to avoid declarer establishing spades and drawing trumps ending in dummy.

Following adages should be mostly confined to when you haven't gathered enough information about the hand yet such that there are still multiple options and you have to guess what to do. Then you follow adages because they tend to reflect percentage actions when there are still lots of unknowns.

I highly recommend the books "How to Defend a Bridge Hand", by Bill Root, and later "Killing Defence at Bridge" by Hugh Kelsey if you wish to improve your game.
Eddie Kantar also has some good defense books, and good quiz books "Kantar for the Defense".

[carmelbobc]

Excellent advice. Thank you.

Question: Might there be a way in which GIB can perform that simulation as soon as S views how a double will be interpreted? Or, give S an option to indicate that he wants the double to be for penalty?

By the way, when I click to "Reply," I do not see a panel in which to do so. So, I'm just adding my reply here.

 Stephen Tu, on 2019-July-15, 15:31, said:

My description above really isn't how the bot decides whether a double is takeout or not. That's in other posts like my response to rustysnow's recent post. It decides whether a double is takeout or not based on its bidding rules database, a collection of thousands of programmer written rules that tell it how to interpret each auction. Given the impossibly huge # of possible competitive auctions, and today's complex norms of which doubles should be takeout vs penalty, these rules only cover a small percentage of auctions, and have massive coverage gaps and errors. So especially on complicated rare auctions, double will often have some not very well defined meaning, unclear whether takeout or penalty, and one is at a guess what to do if robot doubles, and also at a guess whether the robot will leave your double in when you want it to, or take it out when you want it to. It's a hard problem that BBO hasn't made much progress on, the robots are just awful compared to good humans in interpreting doubles.

My first post in this thread is more "given that the database correctly defined the double as takeout, why it decided to leave it in for penalties anyway without a trump stack, which is more often right as the auction creeps higher, than it is at the one or two levels".



To get better, you have to try defend by calculation (picturing declarer's hand, counting out the board, playing out the rest of the tricks in your head), not by following adages. On this board, you should know that declarer has 6 trumps (from the opening), and that he will be able to ruff the spades good (spade position has already been revealed) and get there on the third round of trumps if you just play trumps. So the only hope is to shorten dummy's trumps and keep the trump ace to avoid declarer establishing spades and drawing trumps ending in dummy.

Following adages should be mostly confined to when you haven't gathered enough information about the hand yet such that there are still multiple options and you have to guess what to do. Then you follow adages because they tend to reflect percentage actions when there are still lots of unknowns.

I highly recommend the books "How to Defend a Bridge Hand", by Bill Root, and later "Killing Defence at Bridge" by Hugh Kelsey if you wish to improve your game.
Eddie Kantar also has some good defense books, and good quiz books "Kantar for the Defense".

[Stephen Tu]

Don't really understand the question. Obviously double has to be agreed upon beforehand as takeout or penalty; you can't play double as takeout on board 1 then switch to penalty on board 2 without cheating.

With humans it's partnership agreement. With robots it's whatever is programmed into the rules database. And everyone plays double on this auction as takeout; penalty makes little sense.

But as I said on some hands it can be profitable to leave in takeout doubles on average, though doubling opps into game always entails some amount of risk.