[bbstar]

After 20+ years of development GIB still fails to count points and bids like a beginner.

After 1♦-1♠ // 3♠ what is your rebid with this 19 count?


Sadly, on 10 occasions, the Robots reached 3♠ implying that on all cases GIB decided to pass.
This is called "GIB advanced" bidding system.

Full results

[smerriman]

Bizarre. Basic GIB leaps to 7♠, which is also bad since you could easily be missing an ace, but at least works some amount of the time. No idea what glitch would cause it to pass.

Though of course when you say 'in all cases GIB decided to pass', it only decided once, then what happens at other tables is synchronised to match the first decision.

[bbstar]

Thanks for the reply.
The options for GIB on this hand are pass and 7♠. Both look like all-in-gamble mode.

I have no idea how to run the simulation with GIB basic.
Is it possible to know how any previous GIB (v42, v41, etc) versions handle this hand?

Do the GIB developers run any automated tests with sample hands?

smerriman, on 2025-June-22, 15:36, said:

Though of course when you say 'in all cases GIB decided to pass'

Unfortunately, I don't have the info that the bidding was identical in 10 tables. BBO only shows the final contract of all tables and the bidding of 3 tables.

[smerriman]

South had 0 HCP. Unless somebody did something very stupid, every human would have passed. This guarantees an identical auction at all 10 tables, due to the seeding.

It might be that if you gave this hand to the current version of GIB another 99 times, it would bid to slam (perhaps via slower routes) on all 99. It's just that in a tournament, if GIB comes up with a bizarre play for whatever reason, that play is cached and immediately copied to all other tables who make the same request, giving the appearance of a repeated mistake, whereas in reality GIB was only asked what to do once.

But there may well be something else going on here where passing is more than just a 1 in a 100 glitch. Perhaps the only bids it is allowed to simulate are pass, 4♣, 4NT, and 7♠, because it is too strong to even contemplate 4♠ or 6♠, but when it extrapolates the results of the middle two options, concludes it will always end in 7♠, so chooses pass if grand doesn't make enough times in its sample. Who knows.

The only version I have access to is v21 from early 2012, where it comes up with this at MPs:

1♦ - 1♠
3♠ - 4♣
5♠ - 6♠

But it played 4♣ as a help-suit slam try back then, so quite different.. and at IMPs it sometimes bids 4N over 4♣, and after going through 5NT, East ends up taking a stab at grand.

[johnu]

Good stop at 3♠. If North is void in clubs with 3 spades, and South is void in diamonds, then club ruff, diamond ruff, club ruff, diamond ruff, club ruff and even 3♠ is in danger. Players underestimate the value of getting to a solid part score contract.

[bbstar]

Thanks again for the explanations.

Quote

It's just that in a tournament, if GIB comes up with a bizarre play for whatever reason, that play is cached and immediately copied to all other tables who make the same request, giving the appearance of a repeated mistake, whereas in reality GIB was only asked what to do once.

I expect the caching to be used even if the play was not bizarre at the first place.
The caching (especially in tournament) makes sense.

In this case it might be useful to rebid the hand again with the same version of GIB (with caching disabled). But unfortunately I don't have access to that either.