[pescetom]

Sure johnu would approve the play of GiB at trick 2 here ?

[smerriman]

All blame to South here for their shocking 2♦ overcall.. GIB doesn't read into choices of the second diamond card (in particular whether the T denies the J), so it's basically guaranteed South has 6 and playing the K would usually throw away a trick.

[johnu]

pescetom, on 2023-September-27, 14:31, said:

Sure johnu would approve the play of GiB at trick 2

Nope, GIB North should toss the ♦K at trick one so that South will have no illusions that there will be a logical defense to the hand later on.

[pescetom]

smerriman, on 2023-September-27, 15:26, said:

All blame to South here for their shocking 2♦ overcall.. GIB doesn't read into choices of the second diamond card (in particular whether the T denies the J), so it's basically guaranteed South has 6 and playing the K would usually throw away a trick.

Somehow I guessed you would say that
The overcall is a clear overbid (cards so far were boring and the soccer wasn't going well either), but GiB does describe it as 5+ cards. I wouldn't complain if it hung me for it later in the auction, but I don't see that it should blindly assume I have 6 or that opponents are not overbidding.
"Shocking" I would reserve for GiB not reading into choices of the second diamond card... it's not even a question of having a convention or not, it is simple logic that if South had the J he would show it (why force partner to guess when there is no risk in showing).

[smerriman]

It's not that it doesn't believe 5 card overcalls exist, but the only way you're getting even remotely close to everyone having their bid requires the extra distribution. And even if it included some extreme cases where you did have 5, they'd still be outweighed by all the others. This one is solely down to it not assuming you denied the jack; if that assumption is not in place, low is clearly right.

You know how badly broken GIB is by taking it as a fact wherever possible that everyone has their bid.. can you imagine how much more badly broken it would be if it also looked at every card you played mid-hand and took it as a fact that it was accurately carded?

[pescetom]

smerriman, on 2023-September-28, 14:47, said:

You know how badly broken GIB is by taking it as a fact wherever possible that everyone has their bid.. can you imagine how much more badly broken it would be if it also looked at every card you played mid-hand and took it as a fact that it was accurately carded?

It doesn't (as I understand things) know much about normal carding, but in this case the carding is surely implicit, without any concept of "normal" or "agreed"? It can cost nothing to show the J, which will resolve any doubt partner could possibly have. That doesn't require looking at every card played mid-hand, just a bare minimum of logic. If he shows the J he has it, if he doesn't then Declarer has it.

[smerriman]

pescetom, on 2023-September-28, 15:00, said:

That doesn't require looking at every card played mid-hand, just a bare minimum of logic. If he shows the J he has it, if he doesn't then Declarer has it.

How would you define the logic? If you play x at any point during the hand, then you don't have x+1?

[pescetom]

smerriman, on 2023-September-28, 15:04, said:

How would you define the logic? If you play x at any point during the hand, then you don't have x+1?

Not necesarily: this is a specific situation in which case playing x has to deny by natural logic that you hold x+1 (and thus started with 5).
It should be obvious to the programmer, as it would be to an I/A player.

[smerriman]

But that's exactly what I mean; if this is a "specific situation", then it's no longer as simple as it sounds; you have to come up with the logic for what is a specific situation and what isn't.

"It can't cost" is easy to phrase in human terms, but usually quite complex from an algorithmic perspective - many of GIB's bugs stem from it believing "it can't cost"! I'm guessing you're more referring to general logic about playing cards which don't put partner to an unnecessary guess, rather than specifics about one card denying another - e.g., regardless of carding, if partner held the J, they would have known that playing it would avoid me ever playing the wrong card now.

Anything that involves 'would have known' is simple for humans but highly dangerous territory for robots.. in that case you need to look at all cards you can play, then consider the play from partner's perspective; and no longer double dummy, because double dummy none of the options would have costed.. so now you get into recursive situations where you have to run single dummy analysis for other players for every move you can make, which is an exponential increase in thinking time and immediately too slow to be usable.

It really does seem basic but the more you think about it, the more complex it becomes.. even something as basic as x denying x+1 which is more codable would probably cause more problems than it solves.