[uva72uva72]

My link

IMPs, ACBL robot individual

Some time ago I documented what I thought was a bug in the robot's responses to take-out doubles, namely that it often seemed to bid the weaker of two equal-length suits. Another user kindly explained that we weaker players weren't aware of the wisdom of bidding the weaker suit first if it allowed the robot to bid the second suit conveniently later in the auction. I admit I was skeptical at first, because the robot never seemed to bid the second (better) suit and I always ended up playing an inferior contract in the first-bid suit. But now I get it! I see the brilliance of bidding the weaker (and shorter) ♦ suit first with North's (limited) hand so that it could introduce its ♥s later in the auction. I'm sure Pierre Alabarran is applauding furiously from his grave.

And how about that robot? A free response to a take-out double, at the 3-level, on a 3-count! Che uomo (sort of)!

[mgoetze]

Brilliant bidding, too bad declarer did not take advantage of the 2 in 4 chance of selecting a winning club card on trick 1!

[Stephen Tu]

uva72uva72, on 2014-December-23, 14:56, said:

Some time ago I documented what I thought was a bug in the robot's responses to take-out doubles, namely that it often seemed to bid the weaker of two equal-length suits. Another user kindly explained that we weaker players weren't aware of the wisdom of bidding the weaker suit first if it allowed the robot to bid the second suit conveniently later in the auction. I admit I was skeptical at first, because the robot never seemed to bid the second (better) suit

I think the argument was made to bid *the higher ranking* suit first, in order to bid the second lower ranking suit later if the opportunity arose, while allowing doubler to correct back to the first on the same level. Regardless of whether the higher ranking suit is of lesser or stronger quality than the lower. No one argued to always choose the weaker suit first. Not bidding weakest suit first. Nor bidding strongest suit first. Bid *higher ranking* suit, unless there is a better call appropriate and available (cue bid, scrambling 2nt which I don't think GIB knows how to use).

This auction was of course ridiculous. GIB needs to prioritize bidding majors ahead of minors in response to takeout doubles even when the minor is longer. Bidding the minor first when it is shorter is insane.

[uva72uva72]

As noted, I'm not claiming to be one of the stronger players. I do note that a number of players inserted the ♣10 at trick 1 and all still failed. As defended, it failed because of the 4-1 trump split; but that line would also have failed if trumps were splitting because West upon winning the ♣K could play a ♥ and East could continue them when in with a ♦, so no ♦ ruff in dummy. My conclusion was that I needed 3-2 ♥ and I needed to ruff 2♦s in dummy to get home. That meant I couldn't tolerate a trump lead from West at trick 2, which meant winning the first trick, which meant playing the 10 would only be right if East held both honors. Of course, East could still beat me by playing a trump when in with a ♦, but the opponents might miss one chance to lead a trump; they certainly wouldn't miss two. With trumps 4-1, it all seems moot.

Even if I played a low ♣ from dummy at trick 1, West could switch to a trump and I would be limited to 4 ♥ in North, 3 ♣ in South, 1 ♦ and 1 ♠. And with the 4-1 break I think the contract fails with a ♠ shift as well. But that's as far as my analysis gets me.

[barmar]

This was a statistical anomaly of the simulations.

The book bid is Pass. In 10 out of 11 of my test simulations, it either passed or bid 3♥, evenly split between them. But 1 of them bid 3♦, and this also happened in the live game. In most of the hands it dealt in this simulation, it didn't matter whether it bid 3♦ or 3♥ first, it ended up in the same contract (either 4♥ or something by EW). But there were a couple of hands where 3♦ managed to work out better: there was a red double fit, and it was able to ruff spades with low trumps while using doubler's high trumps to pull the opponent's trumps. If it doesn't matter what you do in 95% of the cases, it ends up making its decision based on a couple of outliers.

This is a known limitation of Monte Carlo methods. We would need to deal many more hands in our simulations to prevent these occasional flukes.

[johnu]

Hmmm, I guess uva72uva72 should consider himself lucky that GIB didn't have a better hand and make the inferential jump to 6♦

[1eyedjack]

uva72uva72, on 2014-December-23, 16:18, said:

Even if I played a low ♣ from dummy at trick 1, West could switch to a trump and I would be limited to 4 ♥ in North, 3 ♣ in South, 1 ♦ and 1 ♠. And with the 4-1 break I think the contract fails with a ♠ shift as well. But that's as far as my analysis gets me.

West is kind of squeezed in S an D on the run of trumps and Clubs, so either South's S is established or North's D. You have to do it without the count, and after eliminating the Clubs throw East in with D (earlier he split them on D lead from N) and E then has to lead D2 to let 6 win in North at t13. Not exactly a single dummy line.