UCI Engine for Komodo 8 and Stockfish 5 spewing odd scores for mainlines.

This post is directed for experienced UCI chess engine users.

I've noticed a weird scoring anamoly whenever I attempt to plug in a position into an engine's UCI application. Whenever I compare an engine's Kibitzer score to the Score CP of the UCI, I always get polar opposite results. I've already tried this method with Komodo 8 and Stockfish 5.

Here I told Stockfish 5 through the UCI:

position fen rnbqkbnr/pppppppp/8/8/6P1/8/PPPPPP1P/RNBQKBNR b KQkq g3 0 1
go depth 10

Which should give me the best possible mainline for the worst opener by white (g4). If you notice in the picture, it scored the mainline as "score cp 76" which makes no sense since both d4 & e4 yield negative results and that would indicate their inferior moves (which would completely disagree with known chess theory).

Here is what Komodo 8 thinks of d4.

Try letting the engines run longer; depth 11 is usually junk.

If you check the UCI spec, I think you'll find that when a UCI engine returns a score value, it's from the engine's point of view. So the signs make sense to me. (You can have GUIs correct the scores to always show the score from White's POV, but this is beyond the scope of the UCI spec.)

@Synaphai - I've let it run all the way to depth-35 for Komodo 8, and depth-40 for Stockfish 5. No matter what, it always returns a negative score. I don't think depth is a concern here.

@EscherehcsE - Yeah, that's the only thing mentioned in the UCI specfication file (the score is from the engine's point of view). If that's the case why does 1.g4 yield a higher score from black's perspective instead of 1.d4 or 1.e4.

Because 1.g4 is much better for Black than either 1.d4 or 1.e4?

I was under the impression that a negative value meant black has the advantage and a positive value meant vice versa for white (similar to the logic found on evaluation graphs).



So what you're both trying to tell me is that the UCI is formatted in such a structure so that a value is not necessarily a raw evaluation score of the overall game but instead a perspective evaluation? (a positive value doesn't mean white is always leading, but instead relates to a turn based mainline)

In the example below I did a mainline analysis of the intiatial starting position (no moving pieces) at a depth of 10:

Stockfish 5 tells me mainline is 1.e4 at a depth of 10 with an evaulation score of 29.

In the next example below, I did a mainline analysis of 1.e4 at the same depth of 10 (it's now deciding what black's mainline will be for 1.e4).

Stockfish 5 decides 1.e4 f3 is most optimal path at a depth of 10 after scanning 61,440 nodes.

My question is:

Just to be sure, what does the second evaluation score mean for white and black? Thanks!

When a UCI engine passes a score to the GUI, it's from the point of view of the engine. In other words, if the engine is evaluating a White move and the score is positive, that means White has the advantage; and if the score is negative, Black has the advantage.

Conversely, if the engine is evaluating a Black move and the score is positive, that means Black has the advantage; and if the score is negative, White has the advantage.

In answer to your question about the second eval score, the engine is actually evaluating White's first move of Ng1-f3, and it's assigning a +36 cp to the move (meaning White has the advantage).

A final comment: Once the engine passes a score to the GUI (from the POV of the engine), the GUI is free to massage the number however it wants and to present it to the user however it wants. Often the user will want it to be presented in the GUI always from White's POV. (That is, a positive score means White has the advantage, and a negative score means Black has the advantage, regardless of whether the evaluation is for a white or black move.)

Got it, thanks. Always good to update your knowledge.

Where can I find Komodo 8 UCI for download?

http://komodochess.com/Komodo8-40a.htm

It's $59.96

(Several older versions are free downloads, though.)

Thank you! On my (older) PC, the older versions perform better.