[johnu]

---

In an ACBL robot duplicate, only 1 other table out of 42 tables got to an excellent 6 diamonds. Every human South took 12 tricks after losing a trick to the Ace of hearts. Every robot North took 11 tricks after a first round finesse of the 9 of clubs in an attempt to make all 13 tricks.

I don't understand the "math" behind GIB's decision to jeopardize a 25 HCP slam contract for a near worthless overtrick and I certainly don't like the line of play. The actual line of play loses to 1/3 of the 5-1 club splits, 40% of the 4-2 club splits, and 80% of the 3-3 club splits where East has the jack or 10 of clubs. How do those numbers add up to make playing for an overtrick even remotely profitable?

I'm not willing to ruff a spade before drawing trump in case east has led a singleton spade and can overruff. Playing as safely as possible, I would win the spade Ace, draw 2 rounds of trump, spade king, ruff a spade, ace and king of clubs, and cross with a trump. East has shown up with 3 spades, 5+ hearts on the bidding, 2 diamonds, and jack small of clubs. When I lead up to the Q9 of clubs in dummy, and west doesn't play the 10, even if I'm 99% sure that east is 3=6=2=2, I just go up with queen and make 12 tricks. If east opened on a 5 card heart suit, however likely, or unlikely, they could be 3=5=2=3 and the club finesse will fail along with the contract.

[johnu]

In this thread, How can GIB possily go down in this slam? there were interesting details about how GIB determines a line of play, although their were multiple lines of play available.

I'm still trying to get a handle on how GIB does simulations and determines a line of play.

I'm assuming when GIB does the double dummy simulation, it only takes the double club finesse when it is working, and only plays clubs from the top when it doesn't work, so the expected tricks are greater when taking the doubble club finesse. Is that right?

But when GIB sees that the double club finesse will produce more tricks on average, doesn't it take the next step, and take the club finesse and simulate what will happen on another sample of hands and see that there's a huge negative expectation?

[barmar]

We're not really sure what's going on here, either. Analyzing GIB's play is difficult.

My thinking is that it assumes the field is in the same contract, so making 6♦ will just score average. Since the defense didn't take their ♥ at trick 1, it has an opportunity for a top by making an overtrick, and it decided the best way to do this was to play West for ♣J10.

Human players can estimate how likely the contract is -- when you're in a thin contract that no one else will bid, you play safe. But I don't think GIB can do this.

At IMP scoring it plays completely differently (although not totally safe, either -- it takes a top ♣ before drawing trumps).