[bid_em_up]

Setting: Money Bridge, 1/2 cent table, vulnerable.

Auction:

1♦-(1♠)-p-(2♠)-
3♦*-(X)-p**-(p)-p

*dumb bot
**dumber me for believing bot has will have some semblance of its explanation of bid.

3D explained as 6+ diamonds, 16+ hcp.

Dumb bot:

void KJ10x KQ9xx AQxx

3♦?? What happened to a normal double of 2♠??

Result: -1100

[arigreen]

Thanks for the report. GIB simulated and judged incorrectly that bidding 3C or 3D would lead to a better score than double. In one of the sims, GIB predicted that a takeout double would lead to partner bidding 3C and going for a number when 3D would have been better. In some of the sims, the double made it easier for the opponents to reach 4S which made. And in some of the sims, bidding 3D directly led to the opponents bidding 3S which went down, when our side would have gone down in a partial had GIB made a takeout double.

There were several results in the simulation where double was the best action: sometimes partner passed for penalty, sometimes we reached the best partial, etc. Overall the decision to bid 3D (or sometimes 3C) over double was very close.

Clearly, the correct approach for this hand would be to simply double without simulating with a hand of this strength and shape.

[allfail]

How does GiB simulate your response after the double? If it simulates it properly it is quite something to know that simulation wise this decision is that close.

[helene_t]

Sounds strange that this is a simulation thing. I suppose your response to a double would also have to be based on simuated so it would be nested simulations. Is that really how it works? I would think that double shows some spades (at least one, lol) and 3♣ would show a strong hand. So it thinks that 3♦ is the smallest lie.

[arigreen]

GIB has certain heuristics for deciding whether or not to simulate. As programmers it is interesting for us to know how close the simulation results are. However this does not indicate in any way that any call other than double is reasonable -- it just means that GIB projects this to be the case.

GIB doesn't use nested simulations -- it would be too slow.