Stockfish only requires twofold repeatition to draw?

The computer analysis of the game here is very confusing, en.lichess.org/ervttYL5

Several times the analysis seems to think a move which would repeat the position once is drawing, even when one side is clearly winning, for example 16.Kf2 is clearly a bad move, but when the position is repeated the computer suggests 18.Kf2 as the best move and suggests it would be a draw. Similarly, many times in the end game you can see spikes for black which suggest draws when all they are doing is repeating a position once, and white is still clearly winning (which the table base actually indicates).

"Stockfish only requires twofold repeatition to draw?"
yes. it assumes that if you repeated once it's because it was the best thing to do

I think you misunderstand the problem here. The built in analysis seems to be treating a twofold repetition as a draw, but obviously a position should be repeated three times before a player may claim a draw. There is nothing wrong with repeating a position once and stockfish should not assume that simply because a player has chosen to repeat a position that it was the best thing to do, it should analyse the position on its merits.

i do understand the problem and i agree with you! but what happens is what happens :o

Definitely seems like a problem with the Analysis. I tested this game and occurring positions in an in-game scenario, and no threefold repetition draw claims were made possible throughout the testing. The bug also happens with local evaluation, and as soon as a move is played, it reevaluates. This indicates that it's a Stockfish bug, and I've seen it discussed elsewhere (external link - www.kaggle.com/c/finding-elo/forums/t/12375/stockfish-doesn-t-understand-draw-by-repetition). This really is a bothersome bug, as it messes up the analysis, and also might recommend a straight up bad move (example being ply 35 in the game you provided, where Stockfish suggests Kf2).

This is indeed a stockfish bug. Debates over this issue have been raging for many years now among stockfish developers. Some people consider it a "bug", while others argue that it is actually a "feature". Removing this "feature" in the past has always decreased engine strength in tournament game play. The reason why this is the case is outside the scope of this discussion, but put simply, assuming 2-fold repetition as draw saves time for the engine when calculating variations as it would assume best play would never result in players going for a repetition unless it is a draw. This debate is literally years old and there is still no consensus. From time to time, stockfish developers try clever ways to remove the "feature" without loss of elo, but they have always failed...

Until now.

Recently, someone managed to do just that.

github.com/official-stockfish/Stockfish/pull/925

Now the debate has re-opened. Will the bug be finally resolved? Time will tell.

Interesting, I suppose the issue here is the difference in requirements between Stockfish being used for competitive play against other AIs and being used to evaluate human games. Thanks for your answers all.

@Toadofsky created a patch for Stockfish. It will probably be deployed on lichess by tomorrow.

@Toadofsky @revoof @cacheyourdreams

Here is another game where the 2fold repetition effect is really annoying.

en.lichess.org/ovHQNa4S/black#0

Both my blunders reported by stockfish appear to be positions where my opponent can simply get back to a previously occured position. Stockfish thinks, "Wait a minute, white is losing, but if white plays this move, it will be a 2fold repetition, which is by definition 0.00! Whoa! This draws the game instantly! Obviously black BLUNDERED by allowing white to get a 2fold rep! BLUNDER ALERT!"

As you can see, it is extremely annoying. I just hope stockfish gets patched asap to avoid this silliness during analysis. :(

#9 FYI I've had to patch this multiple times (when I make a patch, official-stockfish changes something that breaks my patch). Truly a Sisyphean task.

FYI technically 0.00 does NOT mean "draw" but I see that a 0.00 evaluation is not helpful in this situation.

"I just hope stockfish gets patched asap"

I appreciate the sentiment and perhaps you misread #8? Stockfish has been patched, details here if you don't believe @revoof :
http://github.com/ddugovic/Stockfish/commit/fb16fb4ff946f8456c1eded29d903b1273e16b78

Deploying the patched software isn't simple and Lichess' system administration is excellent.

Look, if you want a place to vent about "0.00", here are the Stockfish forums. The Stockfish developers will educate you:
http://groups.google.com/forum/?fromgroups=#!forum/fishcooking