Engine evaluations

Engine evaluations show up in a bunch of places in Chessbook—on individual positions, and aggregated into Soundness. These aren't computed in your browser; we keep a database of pre-computed evaluations covering every position in our openings database, about 63 million of them.

Where are Chessbook's evals from?

All of them come from Stockfish 18, with each position getting exactly 1 million nodes of search. That's about a second of thinking, but a second of modern Stockfish is already far stronger than any human, and openings are the positions engines know best.

The fixed node count is the important part. Soundness compares evaluations along a line—how much you could have kept, compared to how much you kept. If one position was searched deeper than the next, that difference shows up as a concession that isn't really there.

Why don't your evals match Lichess?

We used to take a lot of evaluations from Lichess's cloud eval database. It's a great resource, but not really built for this for a few reasons:

• The evaluations come from lots of different users, running different engine versions and settings
• The depths are all over the place, so adjacent positions can disagree for no chess reason
• A few entries are just wrong—we found some with the sign flipped, scoring a winning position as lost

So now we generate everything ourselves. Cloud evaluations only stick around for rare positions outside the openings database, so a really obscure sideline can still show one—not ideal, but those are positions you'll almost never see.