[uva72uva72]

---

Matchpoints, ACBL robot individual

I can't recall responding to a takeout double in a robot individual previously. Others have had the experience, though, based on their responses on this hand. Of the 23 participants, 3 responded 1♠ with their 14-count (and 3 card suit) and 6 responded ♣. That's quite a few people significantly underbidding their hands and hitting the nail on the head - only those who bid 1♠ and 4 of the 6 who bid 2♣ got plus scores. I played the straight man, bidding 2♥, and got the full Laurel and Hardy - North bypassed its 2 4-card suits to show me its J third in ♣ and the result was not pretty.

I don't recall seeing anything about this in the forums. Has this been documented? North's penchant for bidding 3-card suits seem to be something that BBO is willing to live with (or can't change). But the fact that a significant number of participants know to significantly underbid their hands (including bidding a 3-card suit to stay low) when responding to a t/o double is puzzling, to say the least.

[barmar]

It's probably because of the "best hand" feature. Since West opened, and South also has an opening hand, he knows that North hand is likely to be minimum for his takeout double. And South's square shape suggests staying low. The hand may also be distributional because West opened with a minimum hand, but your flat hand won't be able to handle forces.

Best hand is also why you don't get to respond to takeout doubles very often. Since South always has the most HCP, how often will West and North both have opening hands when West is the dealer?

[1eyedjack]

I considered posting a response concerning why some players went low, but it would detract from the GIB-relevant question of why it would bid 3C on a 3 card suit over the 2H cue.

[barmar]

1eyedjack, on 2015-June-16, 12:30, said:

I considered posting a response concerning why some players went low, but it would detract from the GIB-relevant question of why it would bid 3C on a 3 card suit over the 2H cue.

Yeah, that's a weird result of simulations.

2♠ is the book bid. When it did its simulations, most of the time it didn't matter whether it bid 2♠ or 3♣ -- if they had a spade fit, they eventually found it. But there were a small number of deals where bidding 3♣ allowed it to stop there or in 3♦, which was more successful. There were almost no hands where bidding 2♠ first got it to a better contract. So these exceptional hands in the simulation cause it to bid weirdly.

[Stephen Tu]

It seems to me that it shouldn't be simulating here? If cue bid promises a rebid, as it normally does, to me doubler has to stick to book and not try to guess how the auction is going to proceed. Simulations should only kick in when most of your bids are non-forcing and likely to end the auction so you have to guess what to do now.

[uva72uva72]

My link

On this hand (matchpoints, ACBL robot individual) the robot bid its 3-card ♣ fragment before its 5-card ♦ suit. Simming again? Notice, though, that the notes seem to allow for the fact that North has more ♦s than ♣s even though it bid ♣s first.

[iandayre]

uva72uva72, on 2015-June-16, 21:35, said:

My link

On this hand (matchpoints, ACBL robot individual) the robot bid its 3-card ♣ fragment before its 5-card ♦ suit. Simming again? Notice, though, that the notes seem to allow for the fact that North has more ♦s than ♣s even though it bid ♣s first.

First of all, thanks to Barmar for his comments. Still in all, this is a common type of GIB error that BBO needs to stop making excuses for and start fixing.

[barmar]

iandayre, on 2015-June-17, 14:00, said:

First of all, thanks to Barmar for his comments. Still in all, this is a common type of GIB error that BBO needs to stop making excuses for and start fixing.

Not making an excuse for it, just explaining it. We're discussing the feasibility of a general solution to these kinds of simulation errors, but it's not clear whether the fix would be worse than the disease (there are some situations where it's OK to treat a suit as if it's 1 card longer than it is, and it's hard to enumerate all of them in explicit rules -- simulations fill in the gaps). If there's no general algorithmic fix, all we can do is tweak the rules on a case-by-case basis, and there's only so much time available for this so we have to prioritize.

[iandayre]

OK I'll accept that. I also accept that programming is difficult and tricky. What I would like to see you or other BBO personnel address is the quality of the explanations. I know that Uday and JDonn are, at minimum, solid Flight A players. You may be as well, I don't know. What I do know that the overall quality of the GIB bid explanations is terrible, and that any very good player who reads through them would recognize that. It certainly makes me wonder who wrote them to begin with. But regardless of that, they need to be improved significantly. I know that is not an easy or quick job, but there is no better time to start than now.

[barmar]

I was a solid flight B player (represented my district 2 or 3 times in the NAP, came in 2nd in the district 3 years in a row in the GNT). Then I started working for BBO, which allows me to play ACBL tourneys for free, and I picked up so many masterpoints in robot tourneys (not Lasota rate, but fast for me) that I graduated into flight A (or X in strati-flighted events). Since then my track record in f2f tourneys has been dismal -- I consistently come in at the bottom of Sectional A/X Swiss.

I'm not sure what that has to do with the quality of GIB explanations, though. An important thing to understand is that when GIB makes a bid as a result of simulations, the explanation describes what the bid would have meant if it were the book bid, not the hand GIB actually holds. This is consistent with the Laws on disclosure: you explain your agreements, but players are allowed to deviate.

And even in the case of book bids, we sometimes make the deliberate decision not to include all the details in the description. Often there are a number of rules that lead to the same bid, but most of them are low priority rules that are only used as fallbacks when nothing else is available. The explanation will just cover the most likely rules, to avoid an explanation that's practically meaningless. Remember, if a bid results from multiple rules, GIB explains it as the minimum requirements of all of them, not "explanation of rule 1 or explanation of rule 2 or explanation of rule 3".

I'm not saying there isn't room for improvement, of course there is. But there are thousands of bidding rules, and there are more important improvements that GIB needs than someone going through them all trying to improve the explanations. We try to address this in the natural course of other improvements.

[iandayre]

I'm not talking about GIB's calls matching the description. I am talking about areas where even if GIB's calls matched the descriptions perfectly, there would be problems. Or perhaps even more, when the human player has to make a call and no action comes close to showing the hand. There is a recent example where at one point in the auction, 2H, 3H and 4H all had the same description. You have cue bids that force to some level of THEIR suit. The descriptions of an opening bid of 3NT and that of 2C-2D-3NT are identical - surely there should be some distinction. My posts over the last several months have provided many, many other examples.

The connection between the skill level of BBO personnel, in my mind, and the descriptions is that many descriptions do not derive from solid bridge knowledge and experience. I therefore conclude that they were written by persons with limited bridge experience.

[barmar]

My guess is that at least half the rules (maybe even as much as 75%) are at least a decade old, and we inherited them when we acquired GIB. Everything since then has just been incremental tweaks.

Writing rules for bridge bidding is incredibly hard. Most of human bidding is not done by following clear rules. We learn general principles, and then generalize them, and make inferences. Translating these into concrete steps that a computer can follow in every situation is really difficult.

It's incredibly amazing, IMHO, that it works as well as it does, but there are thousands of rules, and some of them overlap, and there are bound to be lots of glitches. We're fixing them, but we don't have the resources to fix them all (we get dozens of robot reports a day, in addition to the posts here).

[tx10s]

Perhaps someone at BBO can explain how this happened to me (and quite a few others on that hand). I do not have the exact hand layout, but I can describe it well enough to explain the situation. RHO opened 1S, I had 16 HCP and 2 little spades, so I doubled. LHO passed, and my robot "partner" bid 2S, described as forcing. I had pretty much equal support for the other three suits, so I bid 3S, also described as forcing. The robot then bid 4C, described as 4+ clubs and 10+ points. I only had 4C, so I passed. Imagine my surprise when robot showed with only 3 clubs but SIX hearts. Those of us who bid it straight up ended up with poor scores. Those that overcalled 1NT even with no spade stopper were transferred to hearts and only went down one. What was really sad, is if the robot had bid hearts, 4H would have made because the robots could not lead through the robot "partner's" K of spades. I find it hard to believe that any simulation would exclude bidding the 6 card heart suit by the robot in this situation.

[Bbradley62]

When did you have that 4C hand, and in what type of game? I don't see it anywhere in your record since May 31, which is all that is available.