BBO Discussion Forums: Oh dear, Gib - BBO Discussion Forums

smerriman, on 2016-June-05, 02:41, said:

That probably represents how crappy the GIB server is- it often doesn't respond and may have gone to pass as default.

cloa513, on 2016-June-05, 04:56, said:

It wasn't a one-off glitch - identical auction on the other side of the challenge.

smerriman, on 2016-June-05, 05:11, said:

Ah. So, basic GIB, then. Not an excuse, but perhaps a contributory cause. I wonder if advanced GIB would have done the same.

Allowing advanced GIB into challenge matches, even if someone has to pay $0.25, cannot come fast enough, I think. And if the players have already rented advanced bots anyway, they should be allowed in the challenge at no extra charge.

smerriman, on 2016-June-05, 02:41, said:

New to GIB?

1eyedjack, on 2016-June-05, 08:54, said:

My understanding was that the only difference between the two bidding-wise is that the advanced Gib can sometimes stray from the book bid via simulations, meaning this was the book bid.. but I guess Advanced would have been smart enough to overrule itself.

Responses to NT seem to be one of the more clear cut rule-based part of bidding though, which is why I found it so surprising.

smerriman, on 2016-June-05, 14:45, said:

You are mostly correct, but I suspect we have paid less attention to what happens after a 2NT overcall than we have to what happens after a 1NT or 2NT opening bid. I am only mentioning this to explain why perhaps this shouldn't be as surprising as it seems.

Errors like the one in question are usually caused by "gaps" in GIB's rules. For example (I am just making this up), it is possible that there is a rule that tells GIB to use Texas with 6 hearts and up to 12 HCP and to bid Jacoby and then 4H (as a slam try) with 6 hearts and 14-16 HCP. Hands with 6 hearts and 13 HCP would fall into a gap - GIB wouldn't know what to do so, in the absence of simulations, it would go with its default action (which is to Pass).

Gaps are hard to detect unless somebody reports a hand that falls into one of them.

Thanks for doing that. I will make sure our GIB programmers are aware of this particular problem and suggest they take a closer look at followup auctions after a natural 2NT overcall.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

This is a bounch of old problems.
I think the programmers have paid less attention to what happens after a doubling overcall than a 2nt overcall - that's to say they have been not many good Gib's rules to fix its following issues. I am a layman, so here I only offer several example for reference.

- Hand-1
What are Gibs willing to do next after opening 2♠?



- Double says " 3-5♣,3-5♦,3-4♥,2-♠,13+TPs".
Here my question is whether such definition is correct, I think such explanation isn't good because it can't really promise its exact cards on " 3-5♣,3-5♦,3-4♥,2-♠." in the many cases.
Its better definition I think is " takeout, 13+TPs".
But I don't know whether such definition will bring many kinds of difficulties.

- Meanwhile, it seems likely that there are some issues on Gib CC in fact.
After Lebensohl, Gibs have no way to be free to show another 4-card major with invitational hand - that's to say that Gibs can choose to do only one thing, showing hcp/TPs or another major，north responding 3nt is a best example.
Here is some example edited by me.

- Hand-2



- Hand-3


- Hand-4


- Hand-5


- Hand-6


Maybe my nonsense.

johnu, on 2016-June-05, 13:15, said:

Given that he got a response from Fred, I think the OP had a point!

fred, on 2016-June-05, 19:34, said:

I posted this 16 months ago. The exact same bug still exists.

smerriman, on 2017-September-30, 23:49, said:

Gap was closed, because advanced robot doesn't pass: http://tinyurl.com/ybs4qzuq

Basic robot still passes: http://tinyurl.com/y7p399ga

So it's robot not having enough time to think rather than a missing definition.

The advanced robot doesn't pass because a simulation tells it to bid on (obviously any simulation would - and did back then as well). If the gap in the bidding database had been fixed, the basic robot would make the right bid.

Bidding database holes can only be analysed by running basic GIB to see what the "default" bid is (see fred's message above).

(If I'm not mistaken, the 'running out of time to think' issue being discussed in other threads occurs when GIB is simulating, ie only during the play for basic bots - it's not connected to bidding in any way.)

smerriman, on 2017-October-04, 00:19, said:

You could be right, I don't understand this stuff too well. I've passed it to programmers again to check.