[jillybean]

---

About 10 GIB passed this 1♦ opening.

[Bbradley62]

Looks like the same bug that Creeksider recently reported.
http://www.bridgebas...s-a-powerhouse/

[barmar]

Almost certainly. They're both 3=2=3=5 hands. My guess is that it's a hole in the bidding database. It's too strong for 3NT, not enough clubs for 3♣, but the entry for 2♣ mistakenly has an upper limit (probably where the 3♣ bid starts).

[manudude03]

How hard would it be to define pass as 0-4 or 0-5 HCP say?

[barmar]

It still has to find a bid that matches its hand. I suspect there's a default in the program that says "If there's no valid bid, pass."

Although we've seen that GIB sometimes makes bids that don't match the explanation, too. I thought the default was to switch to a simulation.

[xxhong]

IMO, the simulation should never overrule systems. Basically, the gib system has many holes (either undefined bids or redundant bids or bids without logic). Still, the bidding should be based on the system, not the random simulations. IMO, a simulation should only be invoked when there are multiple possible systemic bids to decide which bid is better. This single improvement would make gib much more playable than its current version. Now gib can reject a systemic bid, when the simulation says that a ridiculous non-systemic bid may produce a higher score in a very limited sampling set. Therefore, gib is often seen to break its own system. For example, in strong jumpshift auctions, gib is often seen to jumpshift and rebid its suit with very broken 5 card suit (sometimes Axxxx), which represents a solid 6 cards or longer suit, just because of the favorable simulation results from a very limited sample set. In bridge bidding theory, a lot of conventions are strictly designed with very very few flaws. Therefore, a system abuse is usually the road to disasters and unpleasant experiences playing with gib.

Also, the same is true for defensive plays I believe. A signal system should be respected and the design of the defensive plays should respect that signal system and make plays according to the defensive signals. Therefore, a simulation should be invoked only when there are multiple choices which are not against the current signals. A complete defensive signal may not be easy to develop. However, a simple leading convention, a signal on the first trick and a discard signal on the first discard should be relatively easy to implement and respected.

[barmar]

I think GIB's defensive signals are all count.

[xxhong]

Gib plays no defensive signals at all. Basically, gib's declaring and defense have never been modified by BBO I think. All BBO has done was to improve gib's bidding. The most significant change is the 1NT opening and 4NT RKC. Still, some very basic designs are missing, including a very simple trick count method for declaring, defending, claiming and bidding. The bidding of gib is never based on trick count. It is mostly based on points or high card points and often mess the points and HCP up in NT biddings. Without the basic trick counting, gib rarely makes good decisions at grandslam bidding after RKC.

[barmar]

I'm pretty sure that GIB gives count signals at least 75% of the time. I'm not as sure that it looks at partner's signals.

[manudude03]

I guess these robots just don't like having big hands opposite a 1♦ opener, it passed my 1♦ opener with a 16 count.

Link to hand

[mtvesuvius]

Talk about getting burned for balancing!

[cloa513]

There is a lesson in this for all of us- know your opponents! Even if your both insane bidders it evens out.

[xxhong]

If you play low to discourage and your partner always give low no matter what, that indeed covers a high percentage times when the "signal" and the situation are consistent. That doesn't really mean there is a signal system. Gib basically plays either the lowest card or the highest it doesn't think may cost a trick (here, in many situations, it does cost a trick). That's pretty much all gib plays. I never think Ginsberg wrote anything in defensive carding signals for gib. BBO has never modified the playing part of gib. Gib's defensive plays are largely dependent upon double dummy simulations and it is very poor on the first two tricks because of the huge sample size. There is basically no reasoning, no understanding of the previous plays, no signals, no concrete trick counting and only simulations in a very limited sample size. Actually that's also the major weakness of gib's declaring play. The fate of the contract is often dependent upon the first two tricks.

barmar, on 2011-February-01, 22:41, said:

I'm pretty sure that GIB gives count signals at least 75% of the time. I'm not as sure that it looks at partner's signals.

[barmar]

All I know is that when I making an opening lead of the King from AKxx(x), if partner has 8642 it plays the 4, but if it has 862 it plays the 2.

And when I'm declaring, I can watch its count signals as I play side suits, and they're usually accurate. If I have AKx oppsite Qxxx, and they follow up the line, the suit usually breaks 3-3; if they echo, it breaks 4-2.