lichess.org
Donate

Unlimited, stronger game analysis on lichess.org

be-kind-alwaysedited

Great idea and I hope lichess pushes it! Thank you everyone <3
PS: but it will also depend. I mean, most of us don't really need a higher level of analysis, do we? and this blog explains about how hard it may get to maintain the feature, adding is the easy part https://lichess.org/@/thibault/blog/we-dont-want-all-the-features/q3nOzv4n

Great idea and I hope lichess pushes it! Thank you everyone <3 PS: but it will also depend. I mean, most of us don't really need a higher level of analysis, do we? and this blog explains about how hard it may get to maintain the feature, adding is the easy part https://lichess.org/@/thibault/blog/we-dont-want-all-the-features/q3nOzv4n
gamblej

I have a feeling that blog was written in the twilight hours after a long night of fixing crazyhouse bugs. This philosophy is not a blanket you should throw over any new feature that doesn't personally appeal to you, and I'll try to explain why.

Whereas chess.com can tie up chess professionals and influencers with exclusivity agreements and monetary rewards, Lichess has to work for it. Study is a feature aimed squarely at this influential upper tier of players and analysts, and it is these users who find value in something like a 150 million node search that gives a different result than a 1 million node search on a tricky position.

More generally, if you rely on "we don't want all the features" too much, you can end up with MS Paint while your competitor builds Photoshop. There are a lot of MS paint fans out there who probably don't want it to change. Appeasing that minority in perpetuity is OK for Microsoft because it's a rounding error, but it's not a good survival strategy for a gaming website that must appeal to professionals just to maintain its reach.

So you might say "but you can use their local engine to go as deep on any position as they want". To which I'd say yes, and that's the point. You should be able to do that on the whole mainline without stepping through it like an obstacle course, only to have your evals blown away after reloading (unless you're using Lichess Tools).

I have a feeling that blog was written in the twilight hours after a long night of fixing crazyhouse bugs. This philosophy is not a blanket you should throw over any new feature that doesn't personally appeal to you, and I'll try to explain why. Whereas chess.com can tie up chess professionals and influencers with exclusivity agreements and monetary rewards, Lichess has to work for it. Study is a feature aimed squarely at this influential upper tier of players and analysts, and it is these users who find value in something like a 150 million node search that gives a different result than a 1 million node search on a tricky position. More generally, if you rely on "we don't want all the features" too much, you can end up with MS Paint while your competitor builds Photoshop. There are a lot of MS paint fans out there who probably don't want it to change. Appeasing that minority in perpetuity is OK for Microsoft because it's a rounding error, but it's not a good survival strategy for a gaming website that must appeal to professionals just to maintain its reach. So you might say "but you can use their local engine to go as deep on any position as they want". To which I'd say yes, and that's the point. You should be able to do that on the whole mainline without stepping through it like an obstacle course, only to have your evals blown away after reloading (unless you're using Lichess Tools).
be-kind-always

So you might say "but you can use their local engine to go as deep on any position as they want". To which I'd say yes, and that's the point. You should be able to do that on the whole mainline without stepping through it like an obstacle course, only to have your evals blown away after reloading (unless you're using Lichess Tools).

I'm not against it at all :) I'm just ssaying what I think lichess will consider as the cons of adding it. But in my opinion, the pros overshadow the cons, and lichess should add it.

>So you might say "but you can use their local engine to go as deep on any position as they want". To which I'd say yes, and that's the point. You should be able to do that on the whole mainline without stepping through it like an obstacle course, only to have your evals blown away after reloading (unless you're using Lichess Tools). _I'm_ not against it at all :) I'm just ssaying what I think lichess will consider as the cons of adding it. But in my opinion, the pros overshadow the cons, and lichess should add it.
gamblej

From what I gather, a big objection is they think it will be abused to make corrupt analysis. It's a really odd thing to fear because any contributor in a study can already delete that data, change it around, add obscene comments. Why would someone given write permission on a study craft a special browser extension capable of modifying request data in flight just to add garbage numbers?

I honestly just don't think they've thought it through. And I'm not sure they will unless users make them.

From what I gather, a big objection is they think it will be abused to make corrupt analysis. It's a really odd thing to fear because any contributor in a study can already delete that data, change it around, add obscene comments. Why would someone given write permission on a study craft a special browser extension capable of modifying request data in flight just to add garbage numbers? I honestly just don't think they've thought it through. And I'm not sure they will unless users make them.
TotalNoob69

Building a destructive extension is now something any kid can do. Thibault said that if anything can be abused, it will. Hmm... I guess this could have applied to the kids above... Anyway, I believe such an attack vector already exists now.

But there are technical solutions for this, such as storing the evals for that user only and integrating them into the general cloud with a background process that also has some common sense rules.

Building a destructive extension is now something any kid can do. Thibault said that if anything can be abused, it will. Hmm... I guess this could have applied to the kids above... Anyway, I believe such an attack vector already exists now. But there are technical solutions for this, such as storing the evals for that user only and integrating them into the general cloud with a background process that also has some common sense rules.
gamblej

There's a valid concern about standard game analysis being hackable. In this PR, the way anyone can just replace game analysis by running a higher node count would need to change if someone figures out how to deface games and starts doing it. So yeah, I think it's sensible to not want publishing of local analysis for games, or to restrict it somehow. But people should still be able to run them for personal use (i.e. not publish).

Study is a different ball game. Study should have local analyses and allow publishing by any contributor.

There's a valid concern about standard game analysis being hackable. In this PR, the way anyone can just replace game analysis by running a higher node count would need to change if someone figures out how to deface games and starts doing it. So yeah, I think it's sensible to not want *publishing* of local analysis for games, or to restrict it somehow. But people should still be able to run them for personal use (i.e. not publish). Study is a different ball game. Study should have local analyses and allow publishing by any contributor.
PirateOfAdriaedited

TLDR: Ability to choose engine to run analysis for automatic game annotation would improve Lichess

I opened this discussion before, but in Game analysis forum instead of Feedback. Seems silly now...

Stockfish 11 HCE run in browser is more "in sync" with opening explorer than cloud analysis. I assumed server was running engine with very low depth. I was VERY surprised to learn here in this forum that cloud analysis that we get from 'Request computer analysis' button is 1.000.000 nodes per move. That suggests I highly prefer Stockfish 11. Why even limit Lichess to Stockfish? Berserk 14 and Berserk 4.7 are great engines and are open source. I don't see the downside of more options.

Problem is that Computer analysis engine (Stockfish 18 with 1 million nodes per half-move) is much, much less accurate even when compared to in-browser-engine with low settings. But in-browser-engine can't be used for auto-annotation. I have noticed that now (i think since early June) engine running in browser creates mistake question marks and behaves similarly like Lichess Tools extension (changes if move is mistake/blunder but does not change suggested moves).

Is it that I personally prefer HCE engine evaluation? Or do HCE engines (Hand-Crafted Evaluation or Human-Crafted Evaluation) perform better for evaluation? I know that NNUE (neural network) engines are achieving higher ratings and winning chess engine competitions. Is Stockfish 18 better for titled player level game analysis? Do some players prefer Stockfish 11 and/or Berserk 4.7 over the NNUE engines?

Actually, it would probably be best if we had the option to get annotations based on OPENING EXPLORER, with some HCE engine taking over once there is 20 or fewer games with same position found in explorer.

Using one of chess programs instead is not helpful suggestion, I very much prefer Lichess analysis and I'm sure many others do.

TLDR: Ability to choose engine to run analysis for automatic game annotation would improve Lichess I opened this discussion before, but in Game analysis forum instead of Feedback. Seems silly now... Stockfish 11 HCE run in browser is more "in sync" with opening explorer than cloud analysis. I assumed server was running engine with very low depth. I was VERY surprised to learn here in this forum that cloud analysis that we get from 'Request computer analysis' button is 1.000.000 nodes per move. That suggests I highly prefer Stockfish 11. Why even limit Lichess to Stockfish? Berserk 14 and Berserk 4.7 are great engines and are open source. I don't see the downside of more options. Problem is that Computer analysis engine (Stockfish 18 with 1 million nodes per half-move) is much, much less accurate even when compared to in-browser-engine with low settings. But in-browser-engine can't be used for auto-annotation. I have noticed that now (i think since early June) engine running in browser creates mistake question marks and behaves similarly like Lichess Tools extension (changes if move is mistake/blunder but does not change suggested moves). Is it that I personally prefer HCE engine evaluation? Or do HCE engines (Hand-Crafted Evaluation or Human-Crafted Evaluation) perform better for evaluation? I know that NNUE (neural network) engines are achieving higher ratings and winning chess engine competitions. Is Stockfish 18 better for titled player level game analysis? Do some players prefer Stockfish 11 and/or Berserk 4.7 over the NNUE engines? Actually, it would probably be best if we had the option to get annotations based on OPENING EXPLORER, with some HCE engine taking over once there is 20 or fewer games with same position found in explorer. Using one of chess programs instead is not helpful suggestion, I very much prefer Lichess analysis and I'm sure many others do.