Gib Screws up 3-card ending

[xeno123]

At trick 10, GIB underleads its ♥Ace where it can only cost a trick given the King is out - declarer is known to have two trumps and so only room for a singleton heart.

 

http://tinyurl.com/lhquqpe

[shyams]

At trick 10, GIB underleads its ♥Ace where it can only cost a trick given the King is out - declarer is known to have two trumps and so only room for a singleton heart.

 

http://tinyurl.com/lhquqpe

From West's perspective, it is crazy unlikely for East to bid 4♣ without the ♥K -- especially given the vulnerability.

 

So, "knowing" that East has ♥K, the West Robot IMO believes that "all cards are equal" and leads any heart.

[gwnn]

This is a play that can never gain but will often break even, so technically it can still be a result of Monte Carlo simulations. I think it's a bit much to ask for a fix that can distinguish "works on any distribution" from "works on any distribution simulated". Although I would really like more simulations or I would like to know whether GIB sometimes just gives up on simulating, which seems to be the case.

[xeno123]

This is a play that can never gain but will often break even, so technically it can still be a result of Monte Carlo simulations. I think it's a bit much to ask for a fix that can distinguish "works on any distribution" from "works on any distribution simulated". Although I would really like more simulations or I would like to know whether GIB sometimes just gives up on simulating, which seems to be the case.

 

Yes, I agree that is what happened. I've seen it before, mostly in the context of when a player has "lied" during the bidding.

 

There is a relatively simple fix - near the end of the hand run a handful of simulations with all inferences from bidding turned off. Or even simpler, if two cards are equivalent, play the higher one!