[cencio]

I often play tournaments with the computer as a partner. But often when my companion robot alert, his indications are different from reality. Leading us to declare wrong contracts

[usavref]

cencio, on 2025-August-30, 04:28, said:

I often play tournaments with the computer as a partner. But often when my companion robot alert, his indications are different from reality. Leading us to declare wrong contracts

I have had the same experience multiple times. That is why I will not play in robot tournaments. It's very frustrating.

[pescetom]

cencio, on 2025-August-30, 04:28, said:

I often play tournaments with the computer as a partner. But often when my companion robot alert, his indications are different from reality. Leading us to declare wrong contracts

You shouldn't really be allowed to read its explanation anyway, as this is Unauthorized Information
The idea of bridge is that you should both remember your agreements without help at the table.

But nevertheless this is a huge problem, as the opponents are receiving the same inexact descriptions too, and this is an violation of the laws of bridge.
A corollary is that the descriptions are often the only information available for human partner to learn from, as the official system document is extremely brief and occasionally inexact.

Of course all of this is well known to BBO who do not seem ready even to complete the system document, despite frequent price hikes for the robot.

[benellis58]

Well, if the philosophy is to raise prices while not bothering to improve the GIB robots, system, and definitions, then I suppose that explains many things.

[barmar]

cencio, on 2025-August-30, 04:28, said:

I often play tournaments with the computer as a partner. But often when my companion robot alert, his indications are different from reality. Leading us to declare wrong contracts

Just like human players, sometimes there's no perfect bid for a hand. So it chooses a bid for a similar hand.

Humans do this, too. Most people play that in the auction 1♣-1♥-2♥, the 2♥ bid shows 4-card support. But it's not uncommon to raise with only 3 hearts when other bids would be worse.

[pescetom]

barmar, on 2025-September-02, 12:46, said:

Just like human players, sometimes there's no perfect bid for a hand. So it chooses a bid for a similar hand.

Humans do this, too. Most people play that in the auction 1♣-1♥-2♥, the 2♥ bid shows 4-card support. But it's not uncommon to raise with only 3 hearts when other bids would be worse.

I think that is a poor analogy, if it is not uncommon then honest humans would explain "nominally 4 card but 3 is not uncommon".

I also find it singular that as a representative of BBO you fail to acknowledge or address the issue of system documentation raised here.

[smerriman]

 pescetom, on 2025-August-31, 09:21, said:

But nevertheless this is a huge problem, as the opponents are receiving the same inexact descriptions too, and this is an violation of the laws of bridge.

What is your recommended solution? Even GIB doesn't know the range of hands it may make a bid on, so it's not like improved alerts are an option.

[barmar]

pescetom, on 2025-September-03, 15:31, said:

I think that is a poor analogy, if it is not uncommon then honest humans would explain "nominally 4 card but 3 is not uncommon".

I also find it singular that as a representative of BBO you fail to acknowledge or address the issue of system documentation raised here.

OK, I'll put it more simply.

This comes from the basic design of GIB. GIB is allowed to deviate a little from the criteria of most bids. In particular, the advanced bot uses simulations to decide whether it seems better to make a bid that shows 1 more or less cards than it actually has, or is a point or two different. There are also bids that are not allowed to be chosen or overridden through these simulations, e.g. it should never give an incorrect response to Stayman or Blackwood.

Since almost any bid is essentially an approximation because of this, it would be very confusing if every explanation included +/-1 in its length descriptions. The explanations are accurate the vast majority of the time.

I know it can be frustrating on those occasions when it deviates. Yesterday I had the auction 1D-1H-3D-3H-4H, and we ended up in a 4-2 heart fit because the bot didn't have the expected 5+ hearts.

I play a lot of robot bridge -- 8-10 daylongs/day (even more this month because of the eBridge Cup qualifiers), a bunch of instant tournaments and robot/star challenges, and occasional robot duplicates (I'd play more of these but they don't get much attendance at the times I'm playing). I've seen plenty of goofy bids, but not so much that it's a serious problem IMHO -- at most one or two a day. I'm sure I see more from humans in a typical club game.

For some reason these silly contracts seem to occur more often in "just declare" tournaments -- I think there's at least one in every game. But since everyone is stuck in the same mess, it's a reasonable test of declarer play. Trying to make the best of a bad situation is a skill all players need to cultivate -- many of the example hands in lessons on advanced declarer play techniques (squeezes, endplays, etc.) come from overbidding.

[pescetom]

smerriman, on 2025-September-04, 13:04, said:

What is your recommended solution? Even GIB doesn't know the range of hands it may make a bid on, so it's not like improved alerts are an option.

Are you kidding?
Explain the related agreements, such as they are, with no further embellishments or interpretation of their implications in the actual situation.
That is all the Laws require or allow.

[smerriman]

pescetom, on 2025-September-04, 15:09, said:

Are you kidding?
Explain the related agreements, such as they are, with no further embellishments.
That is all the Laws require or allow.

No, I am not kidding. GIB has no idea what the agreements are, other than a list of thousands of rules, written in complex code. Every decision it makes is a case of going through all of these thousands of rules in a particular order, and testing which ones match and which ones don't.

For example, the logic for a 1♠ opener involves including and excluding hands based on (at least in the old version) exactly 81 rules, each of which consists of numerous conditions and would independently take a while to explain. Would you like to see the giant wall of code that an accurate description of this would be?

On top of that, due to simulations, it has no idea which hands will stray from those rules or how it will stray until it has run a simulation for the specific one it is holding. So, I am seriously asking, how do you propose it explains all of this?

[smerriman]

smerriman, on 2025-September-04, 15:11, said:

Would you like to see the giant wall of code that an accurate description of this would be?

OK, I might have a typo in here somewhere, but here's a rough 'full disclosure' of old GIB's true agreement of what a 1♠ opener is:

Either:

Quote

!((P~.<12~@094)||(P~.<12~@05~.<13~@095.[34])||(P~.<=12~@195431~#18(D)<=7&&#18©<=7~)||(P~.<=12~@195413~#18(D)<=7&&#18©<=7~)||(~#1<0~P~.+#0(2,4)>=#16(7)~@29~.>=32~.~.>=32~.~.>=32~.~.>=32~)||(~C==#1~@05~.+#0(2,13)>=#17(6)&&#3(0,C)==1&&#3(1,C)==1~)||(~C==#1~@05~.[#17(6)-#0(2,13),#17(7)-1-#1(2,13)]&&#3(1,C)<2~)||(~D==#1~@05~.+#0(2,13)>=#17(6)&&#3(0,D)==1&&#3(1,D)==1~)||(~D==#1~@05~.[#17(6)-#0(2,13),#17(7)-1-#1(2,13)]&&#3(1,D)<2~)||(~H==#1~@05~.+#0(2,13)>=#17(6)&&#3(0,H)==1&&#3(1,H)==1~)||(~H==#1~@05~.[#17(6)-#0(2,13),#17(7)-1-#1(2,13)]&&#3(1,H)<2~)||(~S==#1~@05~.+#0(2,13)>=#17(6)&&#3(0,S)==1&&#3(1,S)==1~)||(~S==#1~@05~.[#17(6)-#0(2,13),#17(7)-1-#1(2,13)]&&#3(1,S)<2~)||(~C==#1~@05~.+#0(2,13)>=#17(7)&&#3(1,C)==0~)||(~D==#1~@05~.+#0(2,13)>=#17(7)&&#3(1,D)==0~)||(~H==#1~@05~.+#0(2,13)>=#17(7)&&#3(1,H)==0~)||(~S==#1~@05~.+#0(2,13)>=#17(7)&&#3(1,S)==0~)||(P~.[12,14]~@195422~#18©<=8&&#18(D)>=8~)||(~#11(C,5)~P~.<12~)||(~#11(D,5)~P~.<12~)||(P~.[#8,#9]~@09(5.|4.)..3.[23])||(P~.[#8,#9]~@19[542][542][24][2])||(P~.[25,27]~@09(5[CD]|4.)..3.[23])||(~#11(C,4)~P~.<12~)||(~#11(D,4)~P~.<12~)||(~#11(H,4)~P~.<12~)||(~#11(S,4)~P~.<12~)||(P~.[#6,#7-1]~@09(5.|4.)..3.[23])||(P~.>17~@05~.>22~@09[5-9])||(P~.>17~@05~.>23~)||(P~.>#9~@09(5[CD]|4.)..3.[23])||(P~.[6,10]~@57+[67]D@096D(3.|4[CD])..[1-3])||(P~.[6,10]~@05~.<=12~@57+[5-7]H@096H(3.|4[CD])..[1-3])||(P~.[6,10]~@05~.<=12~@57+[5-7]S@096S(3.|4[CD])..[1-3])||(~#11(C,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(C,3)~P~.<11~@114c[HS]~#18©>11&&#18©<5~@29+~.<16~#c)||(~#11(C,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(C,3)~P~.<11~@114c[HS]~#18©>11&&#18©<5~@29+~.<16~#c)||(~#11(D,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(D,3)~P~.<11~@114c[HS]~#18(D)>11&&#18©<5~@29+~.<16~#c)||(~#11(D,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(D,3)~P~.<11~@114c[HS]~#18(D)>11&&#18©<5~@29+~.<16~#c)||(~#11(H,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(H,3)~P~.<11~@114c[HS]~#18(H)>11&&#18©<5~@29+~.<16~#c)||(~#11(H,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(H,3)~P~.<11~@114c[HS]~#18(H)>11&&#18©<5~@29+~.<16~#c)||(~#11(S,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(S,3)~P~.<11~@114c[HS]~#18(S)>11&&#18©<5~@29+~.<16~#c)||(~#11(S,3)~P~.<11~@11([0-3]|4[CD]))||(~#11(S,3)~P~.<11~@114c[HS]~#18(S)>11&&#18©<5~@29+~.<16~#c)||(P~.>9~@05~.>12~@095C5)||(P~.>9~@05~.>12~@096C[4-6])||(P~.>9~@05~.>12~@097C[3-6])||(P~.>11~@09[6-9]C)||(P~.>10~@05~.>12~@09[6-9]C)||(P~.>9~@05~.>12~@095D5)||(P~.>9~@05~.>12~@096D[4-6])||(P~.>9~@05~.>12~@097D[3-6])||(P~.>11~@09[6-9]D)||(P~.>10~@05~.>12~@09[6-9]D)||(P~.>9~@05~.>12~@095H5)||(P~.>9~@05~.>12~@096H[4-6])||(P~.>9~@05~.>12~@097H[3-6])||(P~.>10~@05~.>12~@09[6-9]H)||(P~.>11~@09[6-9]H))&&(((P~.>9~@05~.>12~@095S5)||(P~.>9~@05~.>12~@096S[4-6])||(P~.>9~@05~.>12~@097S[3-6])||(P~.>10~@05~.>12~@095S5C)||(P~.>11~@095S5C)||(P~.>10~@05~.>12~@09[6-9]S)||(P~.>11~@09[6-9]S))||(!((P~.[#6,#7]~@09(5[CD]|4.)..3.[23])||(P~.[#6,#7-1]~@19[542][542][24][2])||(P~.>11~@095C)||(P~.>11~@095D)||(P~.>11~@05~.<17~@1954~#18©<4&&#18(D)>9~)||(P~.>10~@05~.>12~@095H)||(P~.>11~@095H)) && ((P~.>10~@05~.>12~@095S)||(P~.>11~@095S))))

or the database tells it to bid something else, each with its own set of rules of the same length as above, that bid is not marked as a no-simulate bid (not normally true for openers though), 1♠ is considered a close enough alternative, and a simulation tells it that by bidding 1♠, if its partner strictly assumes it has a *real* 1♠ opener and bids accordingly, then the final contract will lead to a better outcome.

Is that what you're looking for? If not, what? That's all GIB has to work with.

[johnu]

pescetom, on 2025-August-31, 09:21, said:

You shouldn't really be allowed to read its explanation anyway, as this is Unauthorized Information

Lots of things are different with online bridge. You aren't allowed to bid out of turn, make insufficient or illegal bids, or revoke. As far as explanations go, you can ask GIB before the hand, or before the tournament what different bids mean, and GIB won't respond, ever. And as noted later, the GIB system notes are woefully incomplete, and GIB makes a lot of nonstandard bids that almost nobody could correctly interpret. So rather than make decisions almost completely in the blind, it's in the spirit of the game to let the users check the explanations from GIB, especially since many of the explanations don't make sense.