[nige1]

barmar, on 2017-August-31, 10:22, said:

Barmar 24 - Nige1 12 http://webutil.bridg...048256&u=barmar What happened to you on board 15? Was ♥8 on trick 11 a misclick? I guess you also missed a spade pitch, so didn't realize the spades were good before going back to dummy.

Well played, Barmar. Your diagnosis is spot-on. I lost the place and was unsure the ♠s were good. Then I miss-clicked when trying to cash dummy's ♣s. For interest, there's a highly suspect ploy to avoid such debacles. e.g. here: (1) After cashing the top ♠s, claim the rest, to find out if the ♠s are good. (2) Otherwise claim the rest less 1, to avoid miss-clicks. IMO (1) is tantamount to cheating But I reckon (2) is OK if you're as prone to miss-clicks as I am Athough claiming should probably be disabled when playing against robots.

---

Barmar points out that disabling claims is not an acceptable option, Apparently, claiming was introduced into robot-tournaments because many players asked for it.

Normally, the on-line claim rule is simple and fair.: when defenders object to declarer's claim, declarer continues playing until defenders are satisfied, with defenders playing double-dummy.

The software should allow declarer only one claim per deal, however. Otherwise declarer can keep claiming to make up for poor memory or lack of technique.

An alternative remedy for a faulty claim, similar to face-to-face rules, would be for the robot to play the hand out, with a tendency to choose inferior lines -- but that would be more difficult technically and humans would usually complain about the result.

also it's impractical to insist on a human director ruling.

[steve2005]

The first time I ever played with robots I assumed the claim worked the way human tournaments. So I claimed twice not realizing I had actually conceded all the tricks. bummer.
I'm glad BBO changed claims to the current way.

Even so, sometimes when you have a valid claim GIB doesn't always accept it. The case I remember there was 1 trump but can get it easily, or I haven't pull any trump but can handle any bad break once I pull 1 round it accepted.

What your talking about may be technically against the rules but sort of happens in humans. Human may accept when not completely valid and this does happen occasionally. But when you claim opponents see your hand which is sufficient penalty. If Gib sees your hand it should always play perfectly, but Gib isn't programmed to take advantage of this and continues it's usual simulations to guess your hand. If Gib was easy to change it would be a routine modification to let Gib know your hand if you claim.

[steve2005]

How can the claim be decided by a human director? In all robot tournaments there is no director.
How can close decisions be resolved in favour of defenders by Gib. Gib only knows how to play one way that's double dummy. So if it "decides" it will take the right 2-way finesse, do the right squeeze or endplay.

[nige1]

steve2005, on 2017-September-02, 13:51, said:

How can the claim be decided by a human director? In all robot tournaments there is no director.

Agree.

steve2005, on 2017-September-02, 13:51, said:

How can close decisions be resolved in favour of defenders by Gib. Gib only knows how to play one way that's double dummy. So if it "decides" it will take the right 2-way finesse, do the right squeeze or endplay.

Gib doesn't always seem to play double-dummy. Although, admittedly, Gib allows you to perform double-dummy analysis; and it performs double-dummy analysis on distribution guesses during its simulations.

BTW, I think Gib might be right to reject a claim when there's an outstanding trump.

[gszes]

This is harsh but it will indeed eliminate anyone claiming for knowledge purposes. Once a claim is made and the remainder of the tricks are NOT capable of being made either via a (cross) ruff or run NT style a robot should play out the rest of the hand
(double dummy) in the best way possible for the DEFENSE AFTER running all tricks possible via NT style putting off playing the trump suit as long as possible). After getting hammered with a couple of HUGE penalties the claim feature will once again be sparingly used as it should be.

[steve2005]

nige1, on 2017-September-02, 16:25, said:

Agree.

Gib doesn't always seem to play double-dummy. Although, admittedly, Gib allows you to perform double-dummy analysis; and it performs double-dummy analysis on distribution guesses during its simulations.

BTW, I think Gib might be right to reject a claim when there's an outstanding trump.

Yes Gib should reject some valid claims where trump or requires more than playing top cards and certain reject an automatic squeeze etc, The claim method used by Gib I suspect isn't using it's play part of the program at least not exclusively.

Gib usually doesn't know the hand. Sometimes near end there is only 1 possible hand. I am not sure what happens at that point. Either it generates a bunch of hands all exactly the same or it throws the duplicates out and only considers only 1 hand. At this point Gib will play double dummy. When Gib has multiple possible hands all hands are looked at double dummy. Gib has no other way to consider a hand as far as been explained.

Now Gib usually doesn't have the hand right, so it will not always perform double dummy.
Also, there is something wrong with Gib's calculations as it make errors that obviously cost tricks and are not as a result of not knowing exactly what the opponents hand is.
For Example in high NT contract opps led K and dummy has AQJx and Gib played small. There are numerous example of this type of error posted on forum and probably even more not posted. Also, there are more but this example is sufficient to prove there is a major error in Gib programming.

[johnu]

If a claim is declined, GIB opponents should be able to defend double dummy and a GIB declarer should play out the hand single dummy. That should solve the problem of intentional bad claims since having GIB declare hands single dummy should scare any advanced or better declarer.

If human makes a 2nd bad claim in the same tournament or in the next tournament, the claim function should be disabled for the remainder of that tournament, plus 10 more tournaments.

Bad claims should be flagged by BBO software, and players who have more than 3 bad claims in 25 tournaments (as an example) would have the claim button disabled for a specified number of tournaments, say 25 tournaments, or until BBO can review the bad claims.

[barmar]

GIB's claim acceptance logic isn't based on double dummy analysis. It's based on single dummy play given what is known for sure from the play so far (no inferences from bidding, signaling, though). It does assume perfect memory of declarer (suits that someone has shown out of, spot cards that have been played, etc.), and that declarer plays optimally given what he should know. It will never reject a claim just because there's a trump outstanding, if there's a sure way for declarer to draw the trump before it can be used.

Theoretically it can make a false acceptance. As I understand it, it doesn't do a complete analysis, it does a bunch of simulations to see if the defenders can ever prevent you from taking as many tricks you claimed. We have some other software that's supposed to be a perfect claim checker, but haven't done the work to integrate it into BBO. In practice, we don't think the difference is that significant.

There are some good ideas here about how claims should be allowed in BBO robot games. Can't promise anything about if or when we'll implement them.

It's not really feasible to make claims against robots exactly like claims against humans -- there's no way to provide a claim statement that says what order you plan on playing. So GIB is generous: if there's some claim statement that would work, it assumes that's what you're claiming. It's kind of like a game played between champions -- they usually assume the opponents can see the "obvious" line, even if it includes a double squeeze.

[nige1]

The current protocol is so far from face-to-face rules that it might invalidate ACBL competition. After declarer claims, the simplest compromise seems to be:

• The software checks the claim -- if possible simulate single-dummy play by declarer.
• Let's assume that the robot rejects the claim (for whatever reason). Then...
• Let declarer play on single-dummy.
• But let the Robots defend double-dummy.
• Allow declarer no more claims on this deal.