[smerriman]

I'm sure you've seen the thread on BW where claims from respected users are being made that GIB has gotten substantially worse over the last 6 months.

In May Diana stated that:

diana_eva, on 2025-May-07, 09:10, said:

• Robots will now claim in all games when the rest of the play is obvious (expanded from claims only in daylongs and challenges)
• Declarer play by GIB advanced is now a bit better

When asked what changed for the latter point Lorserker then seemed to contradict this, stating that in fact there were no bugfixes made, and the change was solely focused on claims, yet somehow also resulted in a 0.05 IMP / board improvement.

So I ask again, (how) were they 'improved'?

[lorserker]

hi smerriman, are you familiar with the component called "gibson"?

[pescetom]

He's sleeping for once (or unable to get into the forum due to the Oops bug) but yes he is


Please go on.

[lorserker]

we removed gibson in v43. that resulted in a slight improvement in declarer play.
(although the main motivation to remove gibson was to make the robot claim more)

gibson applies only while declaring, only for advanced gib, and only after trick 3.
it has nothing to do with what people complain about.

we'll address the other complaints too, but i won't commit to a timeline, otherwise smerriman will hold me to it

[pescetom]

lorserker, on 2025-September-13, 11:37, said:

we removed gibson in v43. that resulted in a slight improvement in declarer play.
(although the main motivation to remove gibson was to make the robot claim more)

gibson applies only while declaring, only for advanced gib, and only after trick 3.
it has nothing to do with what people complain about.

we'll address the other complaints too, but i won't commit to a timeline, otherwise smerriman will hold me to it

Thanks. I am very surprised that removing gibson would improve declarer play.
But I leave the much more expert smerriman to comment (once he gets over the shock)

[smerriman]

Wow, that's a monumental change.. yes, of course I'm familiar with it, given it made up the bulk of Matt's original paper, which I've read countless times.

Doesn't this reintroduce some of the fundamental flaws with the original version though, namely that it constantly put off decisions like two way finesses, with Monte Carlo telling it it will always guess right later? Or take them too early, thinking that after it leads from hand it will know whether to finesse in dummy.

A classic example is:

---

with spades trumps and drawn, needing 4 tricks, where GIBson takes the 100% line of throwing in the opponents with a diamond, while with this disabled GIB plays a card at random, since all cards are equal double dummy, going down whenever this results it in misguessing hearts.

Very surprising that removing the biggest feature from his work would result in improvement (his original testing showed a marked improvement); it must have been very buggy.

As for holding you to a timeline, no, that is not true in the slightest. Putting this in hidden tag just to separate from algorithmic discussion..

Spoiler

As I've stated before, the issue is not breaking an estimated timeline; that is entirely understandable and to be expected in a lot of cases. And far better than not giving any timeline. The reason I feel very let down by BBO and have become very cynical is due the number of times they've said something to me and others - on the forum, and even in direct emails - and then disappeared forever, without an update, ignoring all requests for a followup years later. A single "sorry, we're still looking but it's taking longer than we thought / we investigated but couldn't replicate / we decided not to go ahead" is 1000 times better than saying nothing.

If you'd like to improve please don't fall into the same pattern

.

[lorserker]

yes, what you say is correct.
but in practice gibson doesn't move the needle.

some (fuzzy) arguments:
- argine doesn't have anything like gibson, and it still plays well as declarer
- even basic gib is a good declarer. almost as good as adv gib, and it never had gibson
- users don't complain about declarer and i didn't notice declarer falling apart since last version either

in spite of the slight improvement when disabling gibson, i don't say that gibson itself does any harm. rather i think that the claim checks kick in more and more often we avoid losing tricks when we could claim instead.

indeed, your example can be solved by a claim check. you can claim by playing a diamond. you can't claim by playing any other card. so the correct play is diamond.
(in practice gib does claim checks only for all tricks. it could do for fewer tricks too, but i didn't like giving up the equity of making N tricks through opps messup when trying to guarantee N-1 tricks. this i will think about again)
i find claim situations simpler than gibson and also frequent enough, for example a subset of bridgemaster problems boil down to proving a claim. (possibly gibson could be better at steering us into a claiming situation - i don't know)

it is possible that gibson will come back in the future, when i dive into it to make it proactively claim, and understand it enough to find any possible bugs in it.
this is distant future, because in the near future we have to take care of bigger issues which users are (rightfully) angry about

[pescetom]

That reads somewhat like "I decided to disable the most advanced part of GiB as the quickest way to make it claim more, which management decided is more important than the tens of major bugs reported on this and other forums"

I hope you are at least right that it makes little difference overall (and therefore Gibson was a huge waste of effort).

[lorserker]

I choose to interpret your comment as good humor, rather than a suggestion that my coworkers and I are not well intentioned.

Communication is not my strong point and I dislike getting into debates on the internet.
Nevertheless, I am present here by reading everything that gets posted about robots, and I try my best to make things better.

For those who like to tinker, the `-B 0` flag disables gibson. I am open to read about your experience and examples you find.

[pescetom]

It was intended to be good humour but dissenting from the BBO choice to strengthen claims rather than fix the bugs. Sorry if it looked like anything different and thanks for your work and attention.

[diana_eva]

pescetom, on 2025-September-14, 08:16, said:

It was intended to be good humour but dissenting from the BBO choice to strengthen claims rather than fix the bugs. Sorry if it looked like anything different and thanks for your work and attention.

The way I understand it, some of the situations where gibson was used are now handled by the robot proactively checking claims. A claim is not just a "button"; it requires the robot to check the play and analyse the position with the information known at that point (for example, "can I claim now?"). That actually strengthens its thinking.


As for why prioritize claims over bugs: claims directly improve the player experience on every single board played. That's why it made sense to focus on this first.