Free online Chess server. Play Chess now in a clean interface. No registration, no ads, no plugin required. Play Chess with the computer, friends or random opponents.
Sign in
  1. Forum
  2. Lichess Feedback
  3. Computer assisted analysis wrong in TB position

The computer assisted analysis is horribly wrong in a particular 6 man ending. It's the evaluation of whites move 65. Bb1 in the first game in this study that is wrong.

65 Bb1 is given double question marks but there is nothing wrong with that move, since it holds the draw according to

Low depth SF can't compare to tablebases. I don't think "request analysis" uses tablebases.

The move holds draw even according to lichess tablebase.
It seems that it's some bug in the analysis engine itself, imo.

[Edit:] Or maybe it's like @lovlas said.

I understand that nothing compares to tablebases, it just struck me as strange error since the surrounding evaluations were 0.0 in both directions, and that SF9 made said move in a rather short TC game.

The reason for asking for computer based analysis is to make sure there were no blunders, so when the analysis says: "there is a blunder right here for you", and that turns out to be false, then the purpose of compter based analysis isn't fulfilled. And in this case the right answer is rather easy to provide. I think the computer based analysis should check the tablebase, 6 man should be practically possible nowadays.

I came across a similar issue of confusion looking at the practice studies; for instance the following position [FEN "8/k7/8/K6p/1P5P/8/8/8 w - -"] was (*) reading as a dead draw on the analysis board, only to switch to #20 after the obvious follow-up of 1.Kb5. In that instance it was due to the limiting factor of the number of lines considered; at the default, of one, it never considers other moves over 1.b5?? - and was simply taking the cloud given line? (*It still gives a drawn position, at least for me, with black to move, if you simply shuffle the king once between the a-b files maintaining opposition).

This is more so a problem in a learning environment, when an accurate evaluation is paramount - though it is unclear to me where the settings to alter the practice engine lie.

Also, and I am unsure if I am alone in seeing this; but, at times, and seemingly at random, the engine will mis-load the position; believing the board to be with the wrong player to move (and often, referring to pieces on the turns prior position, referenced by '--') - whereby it gives fantasy evaluations and lines. [nb. This is from the post-game *.pgn analysis option; either signed in/out; and occurs about once every game in some position, when playing back over the game.]

I don't really know if this fits into your subject. This last game I played had huge difference between initial analysis and full server analysis. Is there a reason for this happening? Getting your game analyzed and needing to do it twice for best results?! That's at least an inaccuracy. :] <

Yes, I experienced the same recently. There are differences, guess it’s simply not in use.

This topic has been archived and can no longer be replied to.