smerriman

Same simple auction to 3NT as gszes. Even looking at the two hands, I'm not sure I'm that unhappy to avoid 6♦, though maybe it has reasonable chances.

I'm surprised you didn't see my post ;)

I would prefer it to be a Soloway jump shift, strong but not showing (just) points but specific hand types. If partner didn't play that, I would expect them to be playing the second most common option according to this poll, which is a weak jump shift; showing *less* than a 1♠ response (definitely not 6+). But playing it as invitational 9-11 is reasonable too. If partner made this bid without discussion, I would expect to be playing on BBO with a weak random partner, and they probably have 6-10 or so.

Advanced robot bids 3♠ at all vulnerabilities for me. So seems to be handled fine?

Nah, that's just GIB playing randomly because South has denied 4 hearts, so it doesn't matter.

It gives a 0.07 IMP boost to 4 from 8642 too, so appears to be that way (but 5 from 8542, so it plays higher of equals I think). Though GIB is horribly inconsistent these days, so who knows. And of course, all of this is nonsensical anyway.

Right click?

Old version of GIB simulated 100 hands, found ♥9 to be better on 5 occasions, and the other 95, ♥9, ♥7, ♥2 were all equal best. It then gave ♥7 a small boost as that's the card you should play if you want to signal you have an even number. That boost (seemingly 0.07 IMPs) outweighed the simulation. Ha.

If you found 40 boards, then your simulations definitely weren't coming from the descriptions. With East holding 11 HCP, and North promising at least 9, it's literally impossible for South to have the values shown. I still get an ace lead every time I try to replicate the situation, though K is very close since it rarely ever costs, and given it can't deal any matching hands for South, it probably includes some random hands where West can get in and give you an extra ruff.

Well, basic GIB at a bidding table bids 5♦, so the book bid is correct. Older GIB also has 5♦ as the book bid. When it attempts to simulate, it can't find a single hand that is remotely close to what it thinks South is showing, thus also makes the book bid of 5♦. That's mainly because it thinks to bid 5 of a minor you need a combined 29 points, and it has promised none, but 1♦ also had a maximum of 22, which is a contradiction. It merges those in the description to exactly 22 points, but clearly still can't understand it. Who knows why your robot strayed from the book bid; maybe it found a single hand which it thought was close enough, and on that hand, 5♣ was fine. Or maybe it crashed while failing to simulate any hands, and ended up passing rather than making the book bid.

As mentioned in the past, the fact it shows a description doesn't mean it's a defined bid. But it may be; will check this later.

I assume the diagram was wrong and the dealer was South, not North, and it's the robot that passed 5♣. And that the robot actually had 13 cards (rather than the 12 shown). I also expect 5♣ will probably be incomprehensible / undefined to the robot, regardless of what the description says.

This wouldn't be a good idea - it would give far too great an advantage to the person playing the challenge second, knowing whether to play safe or not.

I'm sure if you post a link to the hands, you'll get comments.

OK. The part of the OP that you quoted was him getting 57% if he eliminated mistakes, so I think he's talking about 'optimal' play - taking the best percentage line - rather than 'perfect' play - ie what cheaters could achieve by creating multiple accounts to know the hands in advance. Adding the highest score a player achieve on each individual board from my last tournament gives an average of 90%, so while those would go down a bit if I equalled them all, I would expect 70% is a considerable underestimate for 'perfect' play, especially when that hasn't even taken into account some of those scores may not be the top theoretical result. But definitely too high for consistently achievable results.

Have you actually played in the daylong reward tournament being talked about, as opposed to other daylongs? These are a very different beast, being non-best-hand and a better class of players - in the last one I played, only 4 of the 1170 players scored over 70%. I find your numbers very hard to believe. In a best-hand daylong, where you defend rarely and games are slams are common, sure, but averaging over 70% on defensive hands seems impossible to me. Unless by 'best' you're not talking about making the best plays, but making double dummy decisions.

Wow. If that's the worst bid you've seen, you can't have played with the robots very often :) I think you overestimate the number of people who think the robots cheat - most people who play with the robots regularly know they can't possibly cheat due to the number of basic errors they make (and most posts about the robots are about how bad they are; which is also sometimes an exaggeration, but of course they can't both cheat and be bad at the same time). At a table with humans, this would certainly not have been considered a psyche - that's a gross deviation from the agreed meaning, while this one is extremely close (correct number of points, correct quality of the diamond suit - the description says also queen or better - just a single card short). It is also perfectly legal from a bridge perspective - remember that in bridge you do not get to know what the player making the bid holds, but what his partner believes he holds, and it's common for the two to differ even / especially for humans. GIB's robot partner will make all further decisions under the assumption that at least 3 diamonds are held, so no problems there. In human bridge, it only becomes a problem if partner starts learning that you might have 2 diamonds and uses that information in later auctions - GIB never "learns". It also appears to be a reasonable bid; jumping to 4♠ may be better, but if you're going to make a game try, diamonds is where you need the most help. Which just leaves why GIB describes it as 3♦+ but only has 2. The logic for making bids is extremely complex and can't be solely summarised by the points / lengths that are used in bidding descriptions. (For example, of course it won't bid 3♦ with *every* hand that matches that description - many are eliminated by the fact another bid would have been even more suitable, and so on). If every bid was matched perfectly to the description, there would be an extremely large number of situations where not a single bid exists for the robot, and thus it would break entirely. Just like humans, robots have to do the best they can in every bidding situation, even if there aren't any matching "rules". In this case, yes, the robot is programmed that 3♦ should show 3+ diamonds, but there is a lower priority rule that says if you're stuck for a bid, you can also bid 3♦ with 2 diamonds. Should the description therefore change to say 2+ diamonds? No - the rest of the database is programmed under the assumption that you have 3+ diamonds, and that's what should be assumed. If every description was altered to make sure the ranges covered the absolute extreme cases, it's straightforward to see that most of the descriptions would end up completely nonsensical (and considerably more misleading). E.g. would you open Axxxxxx Axxxxx - - 1♠? I wouldn't be upset if you did, but if 1♠ opening were described as 8+ HCP - or even less - then nobody would ever play with the robots.

One thing worth noting is that 'getting near the top scores' and 'playing better' aren't necessarily the same thing. Even if you played every board you receive perfectly, you still may not be able to get anywhere near the top scores in a 16 board tournament, because to make it harder to cheat, not everybody plays the same hands. So you may well receive a bunch of flat hands (where it's impossible to score much more than 50% no matter what you do), while the top scorers received a lot of swingy hands. And of course, if you look at the boards of the top scorers in a specific tournament, you'll see a lot of unusual bids and plays which, when they work, may get a top, but when they fail, may get a bottom - and you're just seeing the ones that got lucky. While making crazy bids does tend to make the robots play equally crazily and can work - and there are definitely some unusual strategies which have a higher than average expectation - most high scorers in the NABC tournaments (where you play far more boards, so random swings tend to average themselves out better) - have tended to say they play generally down the line for the most part. But obviously, the more you get to know GIB's quirks in bidding and play, the more you can start to learn how to work around them and avoid some disasters. There are lots of articles around from these types of players if you search for robot strategies - eg some are linked to in the early comments in this thread.

I don't get it. To satisfy b, it must satisfy i OR ii. It satisfies i, so the fact it doesn't satisfy ii is irrelevant?

Of course, two way is far superior to NMF. But the poll wording was specifically about IF you play NMF, what do the bids here mean, which is exactly what the question here was.

It has no relation to Michaels; it's an Unusual 2NT and GIB correctly describes it as such. Thorvald just put the wrong description in above. Non-vulnerable you can give it Ax.AKQ.AQxx.xxxx and it'll bid 3♣ without simulating. Give it an extra Jack, and it will simulate before leaping to slam. The book bid in the earlier version of GIB was also 3♣, but that was mainly intentional / not a problem - the only specific continuations it defined after U2NT was to give preference to the longer suit (or occasionally bid 3NT); everything else was meant to come from simulations, which makes sense. (There is a weird quirk in the database that tells it not to count a jump to 4M as a book bid in many situations when there are chances of slam, but it's allowed to pick this when simulating, so that's not really a problem.) But they must have really messed some bid priorities / simulation spots up, as back then it would choose from a range of 4-5 different continuations including game in hearts as you'd expect.

GIB doesn't signal. It could have been programmed to signal.. but wasn't. Not much else to say :( Well, other than trying to decide when you should signal, what signal to give, etc, isn't as easy as it sounds. As evidence by the recent Nook AI exhibition, where it totally destroyed WBridge5 by exploiting the fact it signalled too accurately. Not signalling at all is obviously worse, but something GIB players have to learn to deal with.

GIB plays it like the vast majority of BW pollees.

Both are age-old bugs. So a hard no to the subject line.

Hard to say for sure that there is a 'standard' here but without agreements I would agree that pass is the only bid that you can make when you don't want to play game, making 3M slam interest. With agreements you can play 3♣ as Wolff Signoff (forces 3♦ so you can sign off in 3M if you must) or transfers. With more '4♦ inviting game' bids popping up, I think we're past the point where you need to say "Our" in the subject, rather than "Partner's", though if they didn't learn this from the hand a couple of weeks ago..