About the analysis feature

Let me start saying that this is something I REALLY love. It gives you the posibility of learning from your own mistakes, but I have some things to say:

• Sometimes the analysis tells you whether a move is a mistake or an inaccuracy, but if you don't have a somewhat high level you can't see why that move is bad.

  Why does this happen? Because it only tells you the immediatly next move, nothing else. If only the engine showed you four or five moves ahead with the best plays for both sides (which I think it must calculate in order to find the best move) a lot of people would see their mistakes -including me.

• And something I would love: make the engine tell you the only best move in each turn. I know there's the "inaccuracy" flag, but I believe it tells you the move in which you lose 0.5 pawns, meaning that a move in which you lose 0.49 is not flagged and you can't see why that is bad.

Again, amazing feature.

Yeah that first issue is considered a lot here. I can't remember where it was last discussed, but my idea was that any depth of moves deep (shoving the continuation 1 or 3 or 5 etc. Ply long) is incomplete because people are going to question the sub-moves at times, too. Maybe not in all cases, but I think more often than not the deeper analysis lines cause people to ask more questions than they do answer. I guess the idea is that if a single move correction isn't enough, people can paste the FEN of the game into a chess engine and seek into the moves list at the position to get extended analysis from their own engine. It does sound sort of inflexible...it's a hard issue, I think.

As for 49 centi-pawns versus 50 centi-pawns, chess engines are known to be very sensitive to score changes on positional evaluation. A move considered 0.80 points below the best move at a depth of 16 Ply might be considered a whole point ABOVE what was originally thought the best move at 17 or 18 Ply. Even the most powerful engines can be really sensitive, so that's why there is some effort to dissolve the difference between score comparisons.

Well, I understand what you say, but sometimes I have been struggling what the continuation should be after a move and found it after a long time when a simple line stating 3 or 4 moves would have been enough. I also don't have many chances of analyzing the games I play because most of the time I play in Debian and neither Houdini nor Rybka are available, for example, and that's why I found this feature so interesting. Also, I think it would help many players sometimes. For example, it would be nice that when it says "Mate found", it told you what mate it is because some tactics are not easy to see for new players.

Just to clarify... I don't know how the analysis system works, but is the depth of moves showed hard to code? I mean, would it be a pain to change that one move to 4 or 5? I really know nothing.

I agree in the second statement, but since we are always trusting a certain depth analysis I think it would harm no one to trust it always. I mean that if we analyze this with 10 ply (just for example) and trust it to be 0.51, shouldn't we do the same with the best move found with that depth?

Thanks.

I'm going to put this here as an example of what I'm talking about:

http://es.lichess.org/analyse/nednxkba

I was playing with the white pieces. In move 8 he plays 8... d6?! and I play 9. Bf4. It says I've lost almost 1 potencial pawn, but it doesn't how could I have won it.

My move 10. a4 is an inaccuracy (0.4 pawns) but, since it's less than 0.5 pawns it doesn't say what I should have played.

And there are many examples like this in this game. I think it would be really nice to see how could I improve my game by squeezing the positions to the maximum with those little details, not just the big blunders.

The thing about annotations is that as humans we always get an idea of how deep to annotate into move sequence corrections, but this judgement is much more intuitive to attempt to simulate using a computer. Without a variable way of calculating how deep to give the move line corrections, it should be left at a fixed value by default, and my personal preference is 1 because I'm able to think for myself and figure out why that's the better move without having too much of the answer cheated in front of my face.

Yeah...thinking about it more, I can definitely see what you're saying about the second idea. And yeah, the analysis is still prematurely annotated. But, what you don't want to happen is for a "?" or "?!" to always show up every time a move that isn't the best according only to Stockfish is played, because just 0.01-pawn differences are going to smother the move list with "? mistake/inaccuracy" etc. all over. But just listing the best move every time indeed would be a bug fix for this....

I think I can always show the favorite stockfish move, along with the full main line, for every move of the game. It sounds doable.
Why I chose not to do it for this first version is that it makes the PGN huge and completely unreadable for a human. But maybe this is just not an issue.

There's a quote I like: "the winner of a game of chess is the player who makes the second last blunder". It feels very true :-)

I now also like "checkmate is now unavoidable". It's like having a captcha puzzle each time I see it. I'm trying to figure out if it always implies 'blunder'. I'm pretty sure "lost forced checkmate sequence" always means 'blunder'. If "checkmate is now unavoidable" doesn't always mean 'blunder', would it be possible for 'blunder: ' to be prefixed when appropriate?

It was nice trying to figure out why it once recommended a queen sacrifice. This is a wonderful feature :-)

Yes, I don't want the engine to show those moves flagged as inaccuracy, I just want to see the best move. I'm sorry if I didn't express it right.

And, about the main line issue, would it be a possibility to have a some sort of slider or variable to store how many moves you want to see in the analysis? Also, if it's not a problem to show 4 or 5 moves, wouldn't it be always better? I think so because someone who just wants to see one move won't mind seeing more, but someone who wants to see more can't. And I definitely think think that when the engine shows a lost forced mate it should show it right away.

About the PGN feature being readable for a human, is it forced to show the entire analysis? It could only show variations when a mistake or an inaccuracy is made.

Thanks.

Yeah the slider wouldn't change the actual data. When you request a computer analysis the actual replay-analyze link is updated with the annotations as stored on the website; the slider would be a method of hiding extra information, though.

I guess you could attempt to detect how deeply to annotate a move using a few algorithms, but mostly those an engineer of chess software could be familiar with. They generally always just print the full annotation of what the engine is currently thinking at the specified depth, if they print any analysis.

The PGN is only less readable if every single annotation includes 3 or 4 moves deep instead of just 1 move deep. Annotations always can be, and currently are, avoided if the move is not flagged as an inaccuracy or a mistake.

Yes, I think the depth algorithm would be very difficult to implement, but since the data is already there -which I didn't know in the first place- maybe the slider would be a solution to keep the minimalism of the site and also allow players to see more than one move?