[1175]

I started a session of duplicate IMPs, and only got through three hands before I decided to abort and write about them. The first (and by far the simplest) concerns this hand:

---

The bidding here doesn't represent my question, but instead offers a bit of a detour. At 11 tables (including mine), South opened 2♠, and after West passed, North raised to 4♠, and East bid 5♣. At 5 tables (not including mine), South continued with 5♠ (not a winning call - West doubled every time), and at the other 6, East played 5♣ making 6. At 2 tables, South passed initially, then bid 2♠ over a 1♣ opening. North raised to 4♠, and South got to play there (undoubled). At 2 tables, the auction shown occurred. At this point, I will shift to criticizing the Robot's declarer play. While (double dummy) declarer can take all 13 tricks (12 with a diamond lead), one would obviously choose the best play to make the contract, the club finesse.

The bidding shown obviously seems a bit off (North-South have 10 spades, and the auction never contained a jump bid or raise, and the bidding by North-South never reached the number of spades held). Nevertheless, the spade rebid by South suggests six spades, so (firing up my Python script), a total of 2,120,756 hands exist where South has six or seven spades, and between 7 and 11 high card points (North has three or four spades, and between 5 and 9 high card points). Playing to drop the ♣Q succeeds in 1,101,466 of them (51.94 percent), while taking the club finesse succeeds in 1,146,205 of them (54.05 percent). The two extra spaces to hold the ♣Q in the North hand make the finesse the winning play. The Robot played for the drop, and failed by two tricks.

Anyway, my question comes from what happened at the last table. South opened 3♠, North bid 4♠, and East... passed. Why did the robot bid 5♣ after 2♠-pass-4♠, but not after 3♠-pass-4♠? I would feel more likely to want to bid 5♣ after the latter auction.

The truly bizarre bids came from the other two hands, which I will add to this thread a bit later.

[smerriman]

 1175, on 2024-June-18, 22:55, said:

The bidding shown obviously seems a bit off (North-South have 10 spades, and the auction never contained a jump bid or raise, and the bidding by North-South never reached the number of spades held). Nevertheless, the spade rebid by South suggests six spades, so (firing up my Python script), a total of 2,120,756 hands exist where South has six or seven spades, and between 7 and 11 high card points (North has three or four spades, and between 5 and 9 high card points). Playing to drop the ♣Q succeeds in 1,101,466 of them (51.94 percent), while taking the club finesse succeeds in 1,146,205 of them (54.05 percent). The two extra spaces to hold the ♣Q in the North hand make the finesse the winning play. The Robot played for the drop, and failed by two tricks.

Two points to your detour:

a) Those two values are very close. Suppose GIB used exactly the same constraints as you did, samples 100 hands, and calculates how many times the finesse won vs the drop won. The finesse is more likely to have a higher count, causing GIB to take that line.. but with those numbers, around a third of the time the drop will actually come out on top. Those third then become a forum post We don't know how many hands it actually samples, but a BBO rep said once that the paid robot samples "a few dozen", so the free robot would be far less accurate than even that..

b) Remember that GIB is also broken in many ways, and in fact describes 3♠ as nonsensically requiring exactly 12 total points. This totally messes up the simulations to the point where it thinks South is a huge favorite to hold the Q based on points.

 1175, on 2024-June-18, 22:55, said:

Anyway, my question comes from what happened at the last table. South opened 3♠, North bid 4♠, and East... passed. Why did the robot bid 5♣ after 2♠-pass-4♠, but not after 3♠-pass-4♠? I would feel more likely to want to bid 5♣ after the latter auction.

Nothing particularly logical. The database has a section specific to 3 level preempts, which tells it a 5♣ overall requires 20+ total points. That section doesn't apply to 2 level preempts, which fall back to more generic logic that says a 5♣ overall requires 17+ total points.

These rules are all pretty silly but also were almost entirely unimportant when building the database, because it was designed to be used by advanced robots, who will simulate to see whether overcalling 5♣ is better or not than passing. Robots were never intended to make the decision based on points alone; that was just something BBO disabled in order to have something worth paying for.

[mw64ahw]

Transcendence the 12th dimension - the penultimate one before enlightenment.

[1175]

I will include the most "interesting" of the three hands here:

---

Only 15 tables played this hand.

When South stayed out of the auction (8 tables), the bidding proceeded somewhat normally:

---

Not a huge fan of the jump to 6♠ (easy to miss a grand, and on this hand, 6NT makes, while 6♠ goes down with the lead of the ♦A and another diamond), but detailed exploration probably requires tools beyond those available. It surprised me a bit that the Robot bid 3♣ vulnerable versus not, but that doesn't seem worse that South getting into the auction. At 6 tables, declarer took all 13 tricks, at 1 table, South led the ♦A then shifted to a spade (very poor looking at all of the potential discards in the dummy), and South set the contract at 1 table by playing the ♦A and another diamond.

At one table this auction occurred:

---

I see a lot of things I don't like here, starting with East rebidding only 2♠ (I guess the Robot believed that would not get passed out). The 4♦ bid by East shows the ♦A, and East didn't show the void in response to Blackwood, but East-West arrived in a normal contract (down one, as South ed the ♦A and another diamond).

I like my two-suited bids, so the auction at my table shouldn't surprise:

---

Little did I realize the potential danger of my bid (going for a bigger number than a non-vulnerable slam), but obviously East doesn't have much of a desire to leave a double in. I found the diamond lead and continuation, and -450 scored better than any other table, except for the 2 tables that beat the slam.

Not all of the weird bidding came from the Robot:

---

A spade lead allowed this to make all 13 tricks.

Two tables produced this auction:

---

Down eight at one table, down six at the other. First big question: Why did West double 5♦, but not 5♥? I certainly wouldn't want to let North-South play anything over 4♠ undoubled (the auction indicates no fit and bad breaks), and despite the large number of undertricks, the hand scored relatively well for North-South (+3 IMPs for down 6, and -1 IMP for down 8).

One more human oddity:

---

So much for the claim that the Robot interprets all doubles as takeout. I don't know what South intended here, but he obviously did not have "16+ HCP; biddable ♠; 16- total points" (a strange explanation with "16+ HCP" and "16- total points"). Declarer took all 13 tricks, but that (930) still scored more poorly for East-West than slam.

Finally, we get to the table where the Robot took some actions whose reasons I seek enlightenment for:

---

Obviously, I forgive the 3♥ bid's promise of "2+ ♠" (nothing better to bid), but again, 4♦ promises the ♦A. The reason for South's double (on lead) will remain unknown for eternity, but for the second "big" question from this hand, I have to ask: Why did East-West elect to play in a doubled cuebid (a suit that South overcalled, and almost certainly holds five cards in)? On the way to slam, this hardly seems like a contract that will produce the best result. And it didn't. While no defense can beat 4♦, it scored only +510 for East-West (exactly the same as 4♠ making 7), a great score (+5.1 IMPs) for North-South. A human would certainly never prefer to play 4♦X in lieu of 4♠ with the East hand (those spades could become completely worthless). Why didn't West show the ♥K, and why did East leave the double of 4♦ in?

[1175]

 smerriman, on 2024-June-19, 03:48, said:

b) Remember that GIB is also broken in many ways, and in fact describes 3♠ as nonsensically requiring exactly 12 total points. This totally messes up the simulations to the point where it thinks South is a huge favorite to hold the Q based on points.

South has 12 "total points" (if you include 1 for the diamond doubleton, and 2 for the club singleton).

Obviously, one can not draw a statistical conclusion based on a sample size that I could literally count on one hand, but I have watched many hours of the VuGraph over the past couple of months, and I have seen a similar situation come up much more often than I would have expected. In fact, I have a screenshot of one of them that occurred late in a recent tournament.

---

I don't have the rest of the auction (to see more than four lines at a time requires using the scrollbar), but North made whatever Blackwood response the pair uses, and North-South settled in 6♣. West led a diamond, and East took the ♦A and continued diamonds. South started trump by playing the ♣K, and he could no longer make the contract. I would have prioritized distribution over points (particularly at this vulnerability), and take the club finesse through East (easy to say, of course, as a spectator, but as a general rule when I know that one opponent has two or more "open spaces" from the bidding or earlier play than the other opponent, I take the finesse through the player with more room to hold the relevant card). I don't remember the spot card that West led, but I doubt it indicated shortness in diamonds, and singletons (such as club in West's hand) make it more likely for people to bid with fewer points.

[pilowsky]

It's worth remembering that Bridge vs Robots is a video game where the objective is to know more that your opponents about the things it does and doesn't do.

Trying to relate Robot Bridge of any kind to the version played with 4 humans is pointless.

Similarly, asking BBO to 'improve' GIB is like complaining that your car doesn't come with paddles; making it difficult to drive it in a river.

In the end, at best, all you get is wet.

[johnu]

 1175, on 2024-June-19, 05:07, said:

I will include the most "interesting" of the three hands here:

---

The diamond suit does not match your description of 6NT making on a diamond lead.

[smerriman]

 1175, on 2024-June-19, 05:07, said:

The reason for South's double (on lead) will remain unknown for eternity, but for the second "big" question from this hand, I have to ask: Why did East-West elect to play in a doubled cuebid (a suit that South overcalled, and almost certainly holds five cards in)? On the way to slam, this hardly seems like a contract that will produce the best result. And it didn't. While no defense can beat 4♦, it scored only +510 for East-West (exactly the same as 4♠ making 7), a great score (+5.1 IMPs) for North-South. A human would certainly never prefer to play 4♦X in lieu of 4♠ with the East hand (those spades could become completely worthless). Why didn't West show the ♥K, and why did East leave the double of 4♦ in?

I'm assuming your mistake in the diagram was that East actually held the K of diamonds, which makes sense with your story and then matches what GIB actually does.

GIB does not have the slightest concept of how to cue bid towards slam; when partnering with it, you want to avoid such sequences wherever possible.

If you want to know the exact details, after the double of 4♦, its logic is:

- in either seat, if you have first round control in ♦, redouble. Ignore the fact it faked having the ace in an earlier description.
- in either seat, if you have second round control in ♦, and first round control of something else below game, cuebid that
- in direct seat, if you don't have second round control of ♦, bid 4♠ with the QJ of trumps?!? Not sure why this is here, but OK, whatever
- in passout seat, if you don't have second round control of ♦, sign off in 4♠

And as you can see there's a blatant hole that it wasn't told what to do with second round control but no other available bid, which is one of a million bugs relating to the fact it doesn't play Italian-style cuebidding and gets in a complete mess every time.

[1175]

 pilowsky, on 2024-June-19, 22:39, said:

It's worth remembering that Bridge vs Robots is a video game where the objective is to know more that your opponents about the things it does and doesn't do.

Trying to relate Robot Bridge of any kind to the version played with 4 humans is pointless.

Similarly, asking BBO to 'improve' GIB is like complaining that your car doesn't come with paddles; making it difficult to drive it in a river.

In the end, at best, all you get is wet.

I agree with most of that, except for asking BBO to improve GIB. Reducing (or eliminating) human mistakes seems to me like the whole point of writing programs for computers to play games, and while those efforts have resulted in the "solving" of some games, room for improvement still exists with Bridge.

 johnu, on 2024-June-20, 00:12, said:

The diamond suit does not match your description of 6NT making on a diamond lead.

Of course. I entered the hand incorrectly (and it took me three edits to fix it).

 smerriman, on 2024-June-20, 03:50, said:

I'm assuming your mistake in the diagram was that East actually held the K of diamonds, which makes sense with your story and then matches what GIB actually does.

I don't know how I missed that (I usually check hands after entering them). I consider the hand editor a wonderful tools, but obviously, not impervious to human mistakes.

 smerriman, on 2024-June-20, 03:50, said:

GIB does not have the slightest concept of how to cue bid towards slam; when partnering with it, you want to avoid such sequences wherever possible.

That surprises me, as I have actually had a number of successful auctions with it that resulted in slam. I will note your comment, however, as you have far superior knowledge on the topic.

 smerriman, on 2024-June-20, 03:50, said:

If you want to know the exact details, after the double of 4♦, its logic is:...

Thank you for the explanation. Do you actually have notes for explanations of the Robot logic? Do they exist online somewhere? Just curious.

[1175]

I will add the last hand now (which actually has more questionable play elements than bidding, but one of the bids does stand out for me).

The hand (I have checked it this time):

---

Again, only fifteen tables played this hand. East obviously opened 1♦. At one table, South bid 5♣ over that, and the Robot duly punished him for -1100 (although it does seem a bit unlucky that not a single table bid a very high percentage slam).

Seven Souths never got into the auction:

---

At every table with this auction, North led the ♠J, and the Robot chose to take diamond finesse (which seems a bit unusual to me with no opposition bidding), and had 13 tricks without the heart finesse or the fourth spade. The play (which looks slightly suboptimal to me) went win the first spade in hand, heart to the K, spade back to hand, and then take the diamond finesse.

At three tables (including mine) South threw in a 3♣ bid:

---

Obviously, declarer had no trouble with the contract, winning the club lead, playing a diamond to the king, cashing two spades and two hearts, and then taking the diamond finesse for all 13 tricks.

At one table, South made a delayed 3♣bid:

---

This worked out very well for him, as the play went: ♣5-♣A-♣2-♣K (which makes me want to scream "NO!" at my monitor, but declarer still has 13 tricks available through trick 6), ♥ to A, ♠ to K, ♠ to A, ♥ to K, ♠ to Q, ♦ to A, ♦ to K, cash the long ♠, and conceded four club tricks to South.

At one table, this South received an undeserved reward for his bidding:

---

The play began: ♣7-♣K-♣5-♣A (double UGH! - South deserved to have (on a different layout) declarer win the trick with the ♣9). The Robot then played a ♠ to the Q, ♥ to the K, and then a ♥ to the J, which promptly led to down 3. I know you will say "double dummy, declarer still has all 13 tricks, even after playing both club honors at trick one," when I ask "Why?", but still the logic to fix that doesn't appear particularly challenging. At all of the tables (including the next two), only this declarer took the heart finesse (taking it the "wrong" way, presumably placing the ♥Q in South for the 2♣ bid).

Finally, we get to the "questionable" bid which occurred at two tables on this auction:

---

I will finish the play at those two tables before I ask the obvious question. Of course, 4♥ can make 12 tricks (13 with the opening lead of either the ♥Q or the ♥4). At one table, declarer won the club lead with the A, played a ♥ to the A, ♥ back to the K, and a third ♥, North winning the Q. North played a second ♣, declarer winning and drawing the last trump. Cashing the spades now (completing the count on the South hand) becomes the obvious play, but declarer instead played a ♠ to the K, a ♦ back to the K, a ♦to the A, a ♠ back to the A (still 11 tricks available), and then... conceded a diamond (and the rest of the tricks). I won't bother to create a diagram of the three-card end position, but declarer had the good ♠Q (the trick that makes the contract), a small spade (which becomes an overtrick when spades break 3-3), and a small diamond. Play proceeded similarly at the other table with this auction, but this South had discarded too many clubs, and had to give declarer a spade at the end, which allowed the contract to make.

I don't like the 4♥ bid, of course. Why play in a 4-3 (at best) fit, with the bidding suggesting bad breaks (although South only had seven clubs), when an eight-card (at least) diamond fit exists? I can't blame the Pass by East (but I would expect a fifth heart, at least, in the West hand for 4♥). Looking just at the West hand, I also want East to declare this. If North has a singleton club, it will certainly get led, where (suppose East has xx of clubs, and South actually has eight of them) on opening lead, South might not lead the ♣A missing the ♣K.

[smerriman]

1175, on 2024-June-21, 03:43, said:

Thank you for the explanation. Do you actually have notes for explanations of the Robot logic? Do they exist online somewhere? Just curious.

The most up to date version of GIB that is used on BBO, no, though you can do some trial and error at practice tables. BBO refuse to release details about that version of GIB

There was a version of GIB included in the Windows executable ~11 years ago where there is a text file of GIB database rules, though it's the final compiled version and extremely obfuscated.. has been a long term project trying to understand / reverse engineer it. There are a number of differences between that and the current version - the majority turns out very similar, though if I spot differences at a practice table I can sometimes modify some rules to make it more match closely, and then it does give the same results. For reference, some of the main rules relating to what to do after that double of 4♦ are some of the simpler parts of the file, and look like this:

Spoiler

0 QX:X:~(p=0||1)&&(c=#0)>=0&&h!=-1~ 0 . . #1391
1 @QX2 0 . . #1392
1 P:P:~p=1~@QX2 0 . . #1393
0 QX2 0 . . #1394
1d4 (R|e.i.~ei<ac+5*(c<f)||ei=af+5~|P) 99 @479 . #1395
1d3 R 63 @29+~.>=32~#f :H~32~f #1396
1d3 R 63 @09+0#f@29+~.>=32~d[^f] . #1397
1 R:P:~l=af&&a=P&&g=-1~@FORCE2 0 . . #1398
1 R:P:P 99 @479 . #1399
1d3 [+a]#f 62 @09+0#f :d0f #1400
1d3 e.i[^N]~ei<ac+5*(c<f)~ 63 @29+~.[16,32-1]~#f~#22(i)~ ~#8(f,i)~:H~16~f:h~32-1~f #1401
1d3 e.i[^N]~ei<ac+5*(c<f)~ 63 @09+1#i~#22(i)~ . #1402
1d3? e.#c~ec<af+5~ 63 @09+[2-9]#f@29+~.<16~#f@29+~.-#0(1,9+f)>=12~#c :D2f:h~16-1~f:H~12+#0(1,9+f)~c #1403
1d3 e.#c~p==1&&ec<af+5~ 63 @09+[2-9]#f@29+~.<16~#f :D2f:h~16-1~f #1404

Yes, that's a simple part, which may explain in part why they gave up on it

1175, on 2024-June-21, 07:43, said:

I don't like the 4♥ bid, of course. Why play in a 4-3 (at best) fit, with the bidding suggesting bad breaks (although South only had seven clubs), when an eight-card (at least) diamond fit exists?

4♦ shows nothing in GIB's system, since it's the default rebid, so it has no way of knowing there is a diamond fit on rules alone. With no fit and no other rules about what it should do, it falls back to a very basic rule of bidding a new 4 card suit.

"The bidding suggesting bad breaks" and "playing in 4♥ can't be good" is all part of the paid robot's logic, where it thinks about how the auction would proceed after a range of potential bids it can make, and how many tricks it can take in the final contracts it's likely to reach via those methods. It concludes that with no scientific way of getting anywhere better, the best bid is jumping to 6N, seeing that as clearly better than the database's suggestion of 4♥ or any other bids.

[1175]

smerriman, on 2024-June-21, 21:27, said:

Yes, that's a simple part, which may explain in part why they gave up on it

I guess I can understand that.

smerriman, on 2024-June-21, 21:27, said:

4♦ shows nothing in GIB's system, since it's the default rebid, so it has no way of knowing there is a diamond fit on rules alone. With no fit and no other rules about what it should do, it falls back to a very basic rule of bidding a new 4 card suit.

It denies four hearts or four spades, doesn't it? East should probably consider passing with 3-3-4-3 distribution.

smerriman, on 2024-June-21, 21:27, said:

It concludes that with no scientific way of getting anywhere better, the best bid is jumping to 6N, seeing that as clearly better than the database's suggestion of 4♥ or any other bids.

I like that bid (and certainly would consider it holding the West hand).

[1175]

I have another hand to add here that just came up. First of all, don't blame me for any of this. I passed 3♠, East played it there, and my Robot partner misdefended, so the contract only went down by two tricks (a poor score for me).

---

Eleven Souths passed the 3♠ opening bid, which in those cases, became the final contract.

One South bid 4♣ (he played it there, making eleven tricks).

One South bid 5♣, which the Robot raised to slam (making).

The auction shown occurred at three tables (making twice, and going down one once).

The 5♦ bid certainly appears odd from a human perspective, but it seems that the Robot believes that the 5♣ bid by South doesn't show a real suit (which seems strange - why wouldn't South bid his longest suit?). Do I have that right?

[smerriman]

1175, on 2024-June-22, 03:40, said:

It denies four hearts or four spades, doesn't it? East should probably consider passing with 3-3-4-3 distribution.

May well do, but GIB doesn't form the descriptions based on what it has denied by not making other bids, only how the definition for that bid has been programmed.

1175, on 2024-June-22, 03:54, said:

The 5♦ bid certainly appears odd from a human perspective, but it seems that the Robot believes that the 5♣ bid by South doesn't show a real suit (which seems strange - why wouldn't South bid his longest suit?). Do I have that right?

Yep. The logic for how bids are made are specified independently to (and considerably more complex than) the logic for how descriptions are formed. In this case it's told to bid a 5 card suit if it has one, but not to update the description; presumably because it expects other rules to follow will allow bidding a 4 card suit and the description will be picked up from there. But in the very special case of doubling a 3-level preempt, and a 4-level cue, and a 5-level response, nothing else matches, and the description never updates.

[1175]

One of the Robot's continuations on this hand seemed a bit odd to me compared to the continuation on a different response.

---

As usual, I will go through the "non-standard" actions first, and this hand has more than you might imagine. I have concluded that the weakest players show up in the wee hours of the day (UTC) - the time that I played this hand.

One South passed 1♠ (making 6).
One South bid a Jacoby 2NT (the Robot drove to slam, which made).
Two Souths jumped directly to 4♠, (making 6 at one table, and 7 at the other).

At the remaining twelve tables, South made one of the two "obvious" bids, either 2♠ or 3♠. While I didn't really think too much of this hand, I put a high priority on letting partner know that we have nine (or more) spades, so I chose to bid 3♠ (the choice at four other tables, while at the remaining seven tables, South chose 2♠).

When South bid 2♠, the Robot splintered with 4♣, an obvious slam try. Even with a maximum (in HCP), South obviously had nothing else to say other than 4♠, which concluded the auction (making 12 tricks in every case).

At the tables where South made the limit raise, the Robot merely continued to 4♠. Obviously, this "feels" wrong to me (the Robot making a slam try opposite a weaker hand, and one not guaranteed to hold four spades, and not opposite a hand where partner has shown more points and four spades - something as little as JTxx,Axx,Qxx,xxx would produce a very high percentage slam). Of course, on the actual hand, South still can't do anything other than bid 4♠ opposite a presumed 4♣ cue bid, but it still seems like North should do something more than just raise to 4♠. As North, I would probably make two slam tries opposite a limit raise, and it wouldn't bother me a bit that South denied the ♥A.

No doubt some good explanation exists for this behavior (and no doubt you will enlighten me). I await your reply.

[smerriman]

1175, on 2024-June-26, 05:48, said:

No doubt some good explanation exists for this behavior (and no doubt you will enlighten me). I await your reply.

Sure, but it's nothing you haven't already heard; the robot wasn't designed to work with its simulations disabled, but the free robot disables them anyway and blindly follows points-based rules instead.

Over 3♠:

- The free robot follows its generic slam logic of adding total points together; it holds 20 (singletons 2 and doubletons 1, but it subtracts 1 for an honor in a short suit) and you hold a maximum of 12, which isn't enough for slam which requires 33. So it bids 4♠.

- The proper robot is told to consider both 4♠ and possible cue bids, and choose the one that will lead to the best outcome opposite a simulation of possible hands you might hold. It sees that bidding on is clearly best, so bids 4♣ then RKCB no matter what you respond.

Over 2♠:

- The free robot is specifically programmed to splinter with a maximum and shortness, so it does. Even if you respond positively with a cuebid, it will still sign off in 4♠, because now it's back to generic slam logic and you don't have enough combined points.

- Meanwhile, the proper robot is told that it's not allowed to overrule the initial splinter - important for cases where it's revealing information to the human so that it doesn't go crazy (same as e.g. responses to Blackwood), so also splinters. If you bid 4♠, it simulates some hands and concludes it should pass. If you bid 4♥ (showing both the Ace and the upper half of your range in GIB's system), things get interesting; it simulates some hands, but reaches a different conclusion based on the exact set it samples. Assuming IMPs both vulnerable, about three quarters of the time there are enough slams making in its sample that it bids 4NT. The other quarter, it passes. Presumably, if it were left to calculate for longer, it would always bid 4NT.

[1175]

How did the Robot end up in this contract? Well, for one reason, the explanation for the 3♠ bid by West says "4+ ♠; 11-16 total points; forcing" (East doesn't have "13+ HCP" for the 2♠ bid, either). West should probably bid 3♣ instead of 3♠ (assuming it shows clubs), but somebody should bid 3NT (defense can't beat if played by East, and possibly down one if played by West).

---

[1175]

Not quite from the twelfth dimension, but it looks like such a basic mistake:

---

The auction shown occurred at five tables, with declarer going down two tricks at every one of them.

Seven tables played spades (at the two level), making ten tricks in every case, after this auction:

---

At the other four tables, South played either 3♣ or 5♣X, going down (not friendly contracts).

On the auction included with the hand diagram, why did West chose 4♥ an the final contract, when East showed two spades, meaning that East-West have a known eight-card spade fit, and quite likely, only a seven-card heart fit? West doesn't have the values for a 4♥ bid "15-17 HCP; 16-19 total points" anyway.

[smerriman]

Re the latter hand, GIB is totally, absolutely, utterly, ridiculously broken when it comes to penalty doubles.

After North's penalty double of diamonds:

- South's pass of 3♠ promises 3+ spades. If South insteads doubles, it promises exactly 2 spades. If South doesn't have 2 spades, you're generally stuffed; give GIB a 1435 count in this situation and it can't pass or double, so bids 4♦... (at least the free robot will; the proper robot will simulate and invariably choose to lie about its spades.)

- Well, perhaps not so stuffed, because after South "promises" 3 spades, the opponents are also stuffed; any time later in the auction after a penalty double, you're strictly not allowed to bid any suit that the opponents have shown at least 3 cards in.. that rules out 4♠, but I guess with the void counting as 3 points, it think it's too strong to pass..