In this game evalutation chart I noticed a surprise return to zero at move 46:
http://en.lichess.org/sD5naBuM#92Is this because the possibility of draw by repetition was considered by the engine, or is it a bug? Because SCID says about it that it's mate in 6. On a side note, what does 'checkmate is now inevitable' mean? I had a game where my opponent did not succeed to convert this 'inevitable' checkmate.
Position repetition => Draw
That's all there is to it, the rest is the developers of stockfish's problem.
The Stockfish developers discussed fixing it, but they won't because their fix actually results in a tiny (probably in the region of 2 or 3 elo) drop in strength.
Personally, I would take the tiny strength drop, and I'm sure most people would. But their goal seems to be solely to build the strongest engine.
As for your side note... "checkmate is now inevitable" can be taken to mean "inevitable assuming perfect play". All evaluations shown by the computer assume perfect play (at least, in the eyes of the computer).
Thanks for clarifying that part for me. As for the Stockfish engine, isn't it an open-source project? They could create a fork (or release a source patch) with this thing fixed, and the fixed version could be compiled for use on this site. There'd be no practical difference in terms of use cases here, except for correcting the charts, which is a good thing.
Absolutely.
oh I would totally use a fork where repetition is not considered as a draw.
Three-fold repetition did not happen in that game. Two-times repetition is not enough to set evaluation to 0.0 or maybe I am missing something?
Thanks, LM Assios, that's a good reader, but from what claymore said I was able to conclude that the required changes to the code are not so complex. The internal analysis is one thing, but the visual graph is another thing. So if a possibility exists to re-compile the engine with this estimation fixed, it would be preferable.
