Using SunFish for finding principled and/or brilliant moves.

[Siderite]

Hi!

I am the developer of LiChess Tools and I am considering using SunFish.js for "brilliant move" detection. My idea is that if a move looks bad when evaluated by an engine that runs exclusively heuristically on chess principles at a reasonable human depth, like 10-16 ply, but good from Stockfish's evaluation, then it might be considered brilliant.

What do you think?

Thanks!

[foo123]

Hello. The idea sounds solid.
However, as it is ,sunfish.js does not output the score for the move found.

[Siderite]

Thanks for replying! So how does the engine decide which move is better? Any simple way of adding centipawn or winning chances eval?

[foo123]

If you describe how you plan to incorporate and use sunfish.js, then I might come up with some modification that makes it possible.

[Siderite]

Well, probably if I remove all NNUE usage and get a list of candidate moves from a 10-16 ply deep analysis of the position, then a move would be as unexpected as the rank in the list. Perhaps I don't need cp at all. But it would have been nice to be able to chart it. If you have the chart of SunFish over the one from StockFish, then one can see where the paths diverge.

[foo123]

Would a custom cmd "eval" which evaluates all possible moves in current position (up to given depth) suit the use case?

Probably such a command could take long time to complete though..