ACCA PanAm kibitz rule

[CRoberson]

..... I really don't think there is any risk of anybody using kibitz information.

You should rethink that statement. I can tell you how, but not in this
public forum.

[Zach Wegner]

I mean I can think of ways of using it, just not effectively. There are only two ways I can think of: looking at the second move in the PV to try to avoid ponder hits, and using the second move to give you a better move to search. I.e. taking the second move and giving it a bonus or a penalty. The bonus would probably work better for stronger programs (you can trust the move more), and the penalty better for weaker programs. In any case, I don't see how this information would be reliable enough at all. You would have to clear the hashtable (as old searches have uncontaminated entries). I just think it would end up weakening the program overall.

[Tony Thomas]

I mean I can think of ways of using it, just not effectively. There are only two ways I can think of: looking at the second move in the PV to try to avoid ponder hits, and using the second move to give you a better move to search. I.e. taking the second move and giving it a bonus or a penalty. The bonus would probably work better for stronger programs (you can trust the move more), and the penalty better for weaker programs. In any case, I don't see how this information would be reliable enough at all. You would have to clear the hashtable (as old searches have uncontaminated entries). I just think it would end up weakening the program overall.

Also, different program kibitz info's differently. Wouldnt it be almost impossible to detect the kibitzing style of each program and use the info?

[BubbaTough]

Also, different program kibitz info's differently. Wouldnt it be almost impossible to detect the kibitzing style of each program and use the info?

No.

[hgm]

Note that WinBoard and xboard have an -autoKibitz option, which directy forwards the "Thinking Output" of the engine to the ICS.

So any engine that responds to the WinBoard 'post' command, ("Show Thinking" in the WinBoard menu), and diplays its PV info during thinking in the GUI, can meet the kibitz requirement without the need for any modification.

So there really is no excuse at all for not kibitzing.

[bob]

Note that WinBoard and xboard have an -autoKibitz option, which directy forwards the "Thinking Output" of the engine to the ICS.

So any engine that responds to the WinBoard 'post' command, ("Show Thinking" in the WinBoard menu), and diplays its PV info during thinking in the GUI, can meet the kibitz requirement without the need for any modification.

So there really is no excuse at all for not kibitzing.

Does this kibitz just once per move, or every time the engine sends an update? If the latter, it is not so useful as it floods the game with kibitzes...

[hgm]

It only kibitzes the last PV line the engine sent before the move. (Which, for some engines, unfortunately is not the line for the (incomplete?) iteration that delivered the move. But I consider this an engine bug. Some engines even send the PV info of their move during ponder time.

Actually, I hate being overwhelmed with kibitz info in the chat window during ICC tournaments. So I did not even make it an option for the user to send th last N iterations. It is on my to-do list to-do list to create a separate window for kibitz output, to prevent it from cluttering up any discussion you might want to follow.

[bob]

It only kibitzes the last PV line the engine sent before the move. (Which, for some engines, unfortunately is not the line for the (incomplete?) iteration that delivered the move. But I consider this an engine bug. Some engines even send the PV info of their move during ponder time.

Actually, I hate being overwhelmed with kibitz info in the chat window during ICC tournaments. So I did not even make it an option for the user to send th last N iterations. It is on my to-do list to-do list to create a separate window for kibitz output, to prevent it from cluttering up any discussion you might want to follow.

It is an option in Crafty. When playing, I like the "kibitz last PV and score only". When providing real-time analysis of _another_ game in progress, most want to see PV/score as they change, which makes sense...

[sje]

Symbolic will kibitz only a single line and it does so at the end of a search. It should probably be changed to send out one line with the PV and another with everything else.

But if I ever get around to it, instead of a two line kibitz, I'll have the program squirt some HTML to a web server. That way a spectator can see all the good stuff plus diagrams as needed.

[hgm]

But if I ever get around to it, instead of a two line kibitz, I'll have the program squirt some HTML to a web server. That way a spectator can see all the good stuff plus diagrams as needed.

I already added an undocumented option to WinBoard to achieve that. Using that option, WinBoard posts all moves immediately as they are made in a special move file. The browser of those viewing the match downloads this file with moves to update the html page with under JavaScript.

It seems something the GUI should do, not the engine. So that one has access to info from both engines on that web page.

The result you can see here. (This demo uses the moves file from a finished game; There are no games running on that server now. This is also why the moves are played so quickly. Normally the game would be replayed at 1 move/sec until the current position is reached, and after that you would follow the game in real time.)