NCM Stockfish Dev testing

[Jouni]

https://nextchessmove.com/dev-builds

How is +60 ELO to SF8 explained?? Unbalanced book?

[JJJ]

It is compared to Stockfish 7, not 8.That's might explain why there is more elo than it should be.

I also see the time control is lower than Stockfish long time control test. 30 sec +0,3 would explain why more elo.

[Jouni]

They have 30+0,3s and 2 CPU. Should be close to framework 60+0,6s?

[JJJ]

Many patches are doing better at shorter time control.

[Andre]

It's +130 compared to SF7 and around +60 compared to SF8

[JJJ]

It's +130 compared to SF7 and around +60 compared to SF8

Of course it should be +30 compared to SF8 in reality, so I guess time control helps a lot here.

[Jouni]

So they were testing WITHOUT book! And now they are re-testing all versions with 8 moves book - waste of time?

[Jouni]

Now everything is displayed exactly and PGN available, great! But draw after move 34 seems to be a little premature?

[Ozymandias]

PGN available

Not anymore.

[Jouni]

Let see what happens next. But this is definitely bad:

Code: Select all

-draw \
    movenumber=34 \
    movecount=8 \
    score=20 \

My suggestion is to re-test one version per month after SF8. And all 2018 versions.