smerriman

I have a feeling GIBson doesn't take into account IMPs vs MPs, choosing between lines based on which makes the contract more often (Suitplay says ace is best at matchpoints, 9 is best at IMPs). If someone can think of a made up deal which would prove/disprove this, I can test that theory..

Agree that 4NT shouldn't be Blackwood in clubs. But in most situations, you have to read beyond the descriptions when playing with GIB, as thousands of lines of logic that goes into whether a bid can be made can't be summarised down to a few characters.

The rule written in the database here says that if you have 5+ clubs and enough total points for slam to be at least a possibility, then you can bid 4♣, and this is a higher priority than 3N. With less points and clubs stopped, the next highest priority rule will be to bid 3NT. But GIB also has a completely unrelated general rule that says if no other rule in the database matches, then a cheapest new suit at the 4 level shows no extra values and nothing more than a biddable suit. The code takes all possible meanings of 4♣ and outputs the union of the definitions. It just isn't able to determine in this situation, that there isn't a single hand that would result in "no other rule matches". But this is all really a bit moot here, because GIB is allowed to simulate and pick 3NT instead of 4♣, but determines that keeping slam open will get a better score on average. I don't see that this was the wrong choice.

Well, I doubt they did it intentionally. But it results in *GIB* doing it intentionally, yes..

I can't see why I'd want to give up all my usual methods for searching for the best contract just because an opponent doubled at some point.

That's the bidding database, yes.

Nope, nothing like that at all. I can replicate exactly the same thing in the older version of GIB, and the database has a specific rule coded into it that tells it when it can bid 4♦ (nothing to do with controls), and on exactly the same line it says how it should describe the bid (controls).

GIB never understands cuebidding for slam. You can ignore the description entirely; the rule for bidding 4♦ doesn't have anything to do with controls, just whether its total points + partners maximum total points could be enough for slam (as well as a couple of other conditions like sufficiently good trumps).

It wasn't right - most of the times GIB doesn't follow the rules, it has no legitimate reason to do so, and ended up working poorly. It just seems the robot has been programmed to lead a false card a small proportion of the time - that's the only explanation I could come up with. (Note the Nook AI project was able to destroy WBridge5 due to the fact WBridge5 carded too predictably - maybe that's what the developers were trying to avoid, though I'd rather it lead accurately 100% of the time.)

It doesn't. While robots follow lead conventions most of the time (see here; I now think the reason it's not 100% is due to the flag about the weighting for correct signals), they don't make *any* inferences about past plays by either partner or opponents. I.e., it does not reduce the simulated hands by excluding ones where a different lead would have been made. (And equally, when a finesse works, it think it's just as likely that it will fail the next time, even if ducking would have been a 100% losing play).

I think this is the key problem here. When I simulate this, West does bid 4♣ most of the time, though sometimes decides to pass. But in all of the simulations, it's extrapolating a lot of opponent continuations to determine the final outcome. Eg, in the simulation where it came up with pass winning, most of the simulated deals ended with a contract of 4♠ whether you pass OR bid 4♣, because GIB hates passing penalty doubles and it's relying on this when completing the auction. While it did see the double being passed a couple of times for bad scores, it also thought that sometimes passing put the opponents in 3♠, while bidding 4♣ either was going down several tricks, or pushed them into a making spade game.. In one of the deals it even found a bug in the early version of the database (doesn't exist now though), thinking one opponent would cue 4♦ and the other pass it :/

You went wrong in assuming you went wrong. Sometimes you make normal bids, the robots make dumb bids, and you get a bad score because some other humans made dumb bids first.

That's exactly what's happening here. When I get it to simulate from 100 hands, virtually all of them give South a normal 16-17 point hand, and concludes that every card is identical. Three or four of the 100 it assumes South psyched, and that's what primarily breaks the ties. It comes up with different leads every time I try as a result; once it led a heart because one of the psychs was: [hv=handviewer.html?n=sq3hkj7dkt2cak943&w=s985hq52da974c876&e=st4hat96dqj853cq2&d=s&a=1np4np6nppp&v=e&p=dad2d3d6]400|300[/hv] and a heart was the only continuation double dummy says would defeat the contract.

According to the definition, 3NT shows 14-21 points. But in reality it would bid 3NT earlier with up to around 16. So I would it to have at least 17 in this auction. Whether North knows that or not when simulating, I don't know. Why it bids 5♥ instead of 4 of a minor (which my GIB seems to prefer, and it seems to think slam is making quite a lot), I don't know.

It has West holding the ♦A instead of the ♠A. Both are the same from the T/O double perspective, just that the former gives South the stops he promised.

Not really sure what the question is. If it bids 3NT with another hand, would need to know what the other hand was to see what the difference was. [edit]Oops, sorry, I misread and thought 4♥ and 3NT were the other way around. My robot simulates and finds 4♥ a clear favorite with this hand. I guess we can go back to the old argument of a poor set of simulations..

But if he thinks ChCh is within his rights to pause with Kxx, then he wouldn't have complained about pausing with Kx. If he thinks at least one of these is unlawful, the other one seems a much better case. Seems something someone like SB would have deduced, that's all. But this is aside from the ruling at hand.

2♦ used to be new minor forcing as you'd expect. Must have introduced a buggy rule about cue-bidding an opponent's suit that got a higher priority.

No idea how to rule, but why did SB get it wrong? Surely someone of SB's nature would realise that playing East for Kx and then complaining about the unnecessary BIT if East held Kxx basically guarantees the contract?

I knew GIB's redouble actions were messed up, but never quite looked into how much.. .. if you put yourself into West's position in the first auction, and hover over the description of pass of the 2♥ bid.. it promises 3+ hearts for some bizarre reason.. .. and there was a rule [and maybe still is] that you're in the balancing position and the opponents have bid a suit that you and partner have at least 6 cards in, and your maximum combined HCP is not enough for game.. not only can you pass.. but you're *forced* to pass. I sure hope they improved things over the last 10 years and there's a better reason than that these days.

Nobody knows. I recall a very old thread - can't find it now - where one of the GIB developers gave some very basic insights into the simulations (eg saying that GIB includes some hands that don't match the bidding to catch some psychs), but that he wasn't at liberty to reveal any more details. The old Windows version I'm using lets me see what hands the robots simulate when defending though, so I'm trying to reverse engineer the logic a bit.. will let you know what I end up with.

Just a bug in the database. The general idea is that transferring to clubs then bidding 3NT shows extra values, otherwise you'd just bid 3NT directly, so they updated what 3NT "shows" in the database to promise the extra values. But they didn't update the logic part. Generally, GIB would bid 3♠ with a singleton spade after transferring. But there's a rule that says if you have 9-12 HCP, and your singleton is an A or K, bid 3NT, and describe it as showing 13-15. (And it has 13-15 with a singleton king, it goes back to bidding 3♠.. guess they just have things backwards by mistake.)

There's a few references I haven't quite nailed down yet, but yeah, I'm getting there.

Somewhat surprisingly, it does conclude the slam is going down on 73 of the 290 cases I mentioned above, so that's not actually the case. It puts more emphasis on the 4 clubs than the 25 points.

BBO definitely broke this one somewhere along the line, as it never used to do this; ran this one through my version 20 times and all 20 simulations resulted in it pulling to 4♥. Wasn't even close. Not sure exactly what update caused it, but I do remember there was a definite increase in posts on the forum about GIB not pulling doubles like this.