lichess.org
Donate

Engine bug?

Lord_Fesneu

First of all, a massive thank you for lichess which is a fantastic platform!

I have noticed something off in the engine valuations: there are moves that are stronger (according to the engine's own valuation) than the ones suggested by the engine.

Let me show a specific example:

In this position (white to play) the engine suggests Nf3 as the strongest move. The second line suggested is Nc3:

1.png

If I play the engine top move, the valuation is +1.3.

2.png

However, if I play another move: f3, then the valuation becomes +1.8.

3.png

So f3 is an even stronger move, according to the engine, but is not suggested at all by the engine.

I've tried the larger model, more depth, more thinking time, it did not change the suggestion.

I have noticed this on several unrelated occasions. It makes me lose faith in the engine evaluation.

I am probably misunderstanding something, can anyone shed light on what's happening?

Thanks!

First of all, a massive thank you for lichess which is a fantastic platform! I have noticed something off in the engine valuations: there are moves that are stronger (according to the engine's own valuation) than the ones suggested by the engine. Let me show a specific example: In this position (white to play) the engine suggests Nf3 as the strongest move. The second line suggested is Nc3: ![1.png](https://image.lichess1.org/display?fmt=png&h=0&op=resize&path=a0gTChSku-aw.png&w=864&sig=fcbaf5598870e16259cf35bbedbe0727d92968ab) If I play the engine top move, the valuation is +1.3. ![2.png](https://image.lichess1.org/display?fmt=png&h=0&op=resize&path=Z-1PpNJf3ZYS.png&w=864&sig=9cc0b6f99a418669d443051210b5206e2aaac9bb) However, if I play another move: f3, then the valuation becomes +1.8. ![3.png](https://image.lichess1.org/display?fmt=png&h=0&op=resize&path=Yus3rU-G2aoK.png&w=864&sig=79c45aa35a66992b669592844ef355e2b4f20a30) So f3 is an even stronger move, according to the engine, but is not suggested at all by the engine. I've tried the larger model, more depth, more thinking time, it did not change the suggestion. I have noticed this on several unrelated occasions. It makes me lose faith in the engine evaluation. I am probably misunderstanding something, can anyone shed light on what's happening? Thanks!
cooked_impala

I think its because every time you play a new move, stockfish rethinks from the very beginning, resulting in variation (stockfish isnt perfect).

I think its because every time you play a new move, stockfish rethinks from the very beginning, resulting in variation (stockfish isnt perfect).
StevenEmily

the analysis only goes up to a certain depth. So your browser stockfish can get an improved result when it searches farther than the computer evaluation request.

an example is

https://lichess.org/4fmlWM2e/white#0

stockfish evaluation only realizes it's losing after a certain point. However, your browser can eventually figure it out even earlier than the evaluation did.

the analysis only goes up to a certain depth. So your browser stockfish can get an improved result when it searches farther than the computer evaluation request. an example is https://lichess.org/4fmlWM2e/white#0 stockfish evaluation only realizes it's losing after a certain point. However, your browser can eventually figure it out even earlier than the evaluation did.
Toscaniedited

Use multiple engines to get a tree branching view of what is possible.
Some times the database just does not have enough popularity in the move choices.
The popularity might be biased by the ratings of the players that played the move.

If you post the FEN, maybe then you could get the list of moves that 3000+ rated engines would play.

There are a lot of chess engines in the world and some if you give them a bad FEN code will still try to analyse the position. Some cannot even see a mate in 7 and they are rated over 3000.

Test your engines with this and if it does not pass, then fix your engines so they do or report it to their owners:
Mate in 7 FEN: 5rk1/1p6/6p1/1Q3p2/8/4P2R/3PKPP1/8 b - - 0 1
Bad FEN: 5rk1/1p6/6p1/1q2Qp2/8/4P2R/3PKPP1/8 b - - 0 1

Use multiple engines to get a tree branching view of what is possible. Some times the database just does not have enough popularity in the move choices. The popularity might be biased by the ratings of the players that played the move. If you post the FEN, maybe then you could get the list of moves that 3000+ rated engines would play. There are a lot of chess engines in the world and some if you give them a bad FEN code will still try to analyse the position. Some cannot even see a mate in 7 and they are rated over 3000. Test your engines with this and if it does not pass, then fix your engines so they do or report it to their owners: Mate in 7 FEN: 5rk1/1p6/6p1/1Q3p2/8/4P2R/3PKPP1/8 b - - 0 1 Bad FEN: 5rk1/1p6/6p1/1q2Qp2/8/4P2R/3PKPP1/8 b - - 0 1
NraniAnI

so interesting!!!

so interesting!!!
NraniAnI

can we be friends? (I know english a little bit)

can we be friends? (I know english a little bit)
NraniAnI

where is comments?

where is comments?