[pdca]

Here's how it went

When there were 4 tricks to go, GIB somehow "forgot" about the last trump and went down 2 instead of pulling it and going -1.

[Bbradley62]

I'm curious about the 2♦ bid. Does it deny a 4-card major?

[mgoetze]

That is indeed a strange mistake, I can only assume that in every single simulation the ♠7 and the ♥10 were in the same hand and so GIB thought it didn't matter. Was this a basic robot or an advanced robot?

[pdca]

mgoetze, on 2011-June-13, 02:37, said:

That is indeed a strange mistake, I can only assume that in every single simulation the ♠7 and the ♥10 were in the same hand and so GIB thought it didn't matter. Was this a basic robot or an advanced robot?

I'm not sure, happened in robot tournament.

[BunnyGo]

mgoetze, on 2011-June-13, 02:37, said:

That is indeed a strange mistake, I can only assume that in every single simulation the ♠7 and the ♥10 were in the same hand and so GIB thought it didn't matter. Was this a basic robot or an advanced robot?

That was my thought too, but that would be virtually impossible with even 100 simulations.

[barmar]

pdca, on 2011-June-13, 07:06, said:

I'm not sure, happened in robot tournament.

Tournament robots are all advanced. You only get basic robots with the $1/week purchase.

[barmar]

BunnyGo, on 2011-June-13, 07:23, said:

That was my thought too, but that would be virtually impossible with even 100 simulations.

Even with only 20 simulations, it's about a million to one. That's the power of geometric progressions.

[BunnyGo]

barmar, on 2011-June-13, 18:20, said:

Even with only 20 simulations, it's about a million to one. That's the power of geometric progressions.

Yeah, but I said 100 because with people playing with robots as often as they do and as many hands as they do a 1 in a million occurrence isn't too unexpected. Whereas a 1 in a 2^100 occurrence is (almost) always a statistically significant event.

Actually, another possibility is that it gave a 100% likelihood to the same player to have the long spades and hearts, and so all the simulations were based on a faulty premise.

[barmar]

That may be it. It knows the exact spade distribution because North didn't overcall 1♠. And if it interprets North's takeout double as showing heart shortness (ignore the explanation, which says 2+H), then South has to have the long heart.

[cloa513]

It doesn't which hand has the last heart- it must be taken out when you still have ruffing strength and cashable cards.

[barmar]

If the hand with the last heart also has the last spade, it doesn't matter which order you play those two tricks. GIB doesn't understand the concept of a 100% play. If its simulations show that two plays get the same result, it picks one of them at random.

I can't count the number of times I've seen GIB unnecessarily cash a trick in a side suit either before or in the middle of drawing trumps. Luckily, it rarely runs into bad breaks that cause this to be punished.