[pretender]

Vul vs NV, South (human, others GIB) deals and passes.
(1♦)-1♥-(1♠)-4♥
(p)-p-(4♠)-ppp

East (the 4S bidder):
AJ842
4
654
Q874

West:
KT65
J5
AQT973
A

[Stephen Tu]

Whatever is preventing West from bidding 4s should be fixed, then hopefully East won't think a hand as good as West's is possible anymore, then it can pass.

[dwar0123]

Wests pass is as crazy as easts bid of 4♠

[nathan2008]

normal..... don't push GIB.... I even saw I pushed GIB from 4d to 6d, and 6dX= lol.

[barmar]

I agree that West should bid 4♠, although I'm not sure how to translate it into a bidding rule. Maybe we need special rules to deal with bidding after the opponents have put the pressure on with a preempt; currently, GIB just goes through its normal bidding logic: I have 15 total points, partner has shown 6, and 21 points isn't enough for game. East's book bid is also pass, but when it did simulations it decided 4♠ would likely work out on 8 out of 11 simulation runs in IMP mode, although only 3 out of 11 in MP mode.

[Stephen Tu]

West should sim also, and if West thinks it's only worth 15 total points then the evaluator needs to be fixed. 9 cd fit = upgrade, 6 cd suit = upgrade, I think it's worth about 17 total points. Any chance of implementing K&R evaluation? 1/2/3 shortness points only is too conservative. with 9 cd fit it should be more like 1/3/5 if counting shortness only.

And in comp it should prob give partner an 8 count on average, when under pressure.

[barmar]

The hand evaluation function is only given a single hand at a time, I don't think there's an easy way to revise it to take fits and other information from the auction into account. So instead we're forced to handle these kinds of things in the bidding rules and simulations.

The problem is that West's hand isn't close enough to any of the hands that produce a 4♠ bid, so it doesn't include that bid in its simulations. Simulations work by adjusting the given hand up and down by 3 points, and moving a single card from each suit to another suit, then looking for rules that match each of the adjustments. But the rule for bidding 4♠ requires 19 points, so adding 3 points to this hand isn't enough to include it. Change ♠K to A, and it includes 4♠ in the simulations and chooses it.

As you said, we should be more eager to bid game with a 9 card fit. What we could do is add a rule that says only 23 combined points are needed for game with a 9-card fit. Then the simulation would find this rule, and probably choose it.

[Stephen Tu]

Even if not adjusting for partner's and opp bidding, I've seen GIB underbid a lot when holding good long 6 and 7 cd suits. A lot of authors advocate counting long suit points as well as short suit points; adding a couple points for the 6 cd suit would get GIB closer. Then maybe combined with a "24 pts with 9 cd fit" it would be enough.

K&R doesn't take other bidding into account for initial eval. It was developed for computers by Edgar Kaplan & Jeff Reubens, published in Bridge World, designed to mimic Kaplan's judgment. If GIB is using static evaluation, I think it would be an improvement.