Then as I said, one should simply take each fail high position in the tree, and re-search with a relaxed beta bound, to see how many fail low the second time around. Then you _know_ how incorrect this assumption is.

Usually there is just no reason to do this. Root is an exception. We need to ...

In this example, the resolved score won't fail low unless it's <0.45 (I've actually never seen a 0.75++ fail high with Rybka or Naum that fails that badly).


SF's score lives much more than Rybka's and Naum's, because of different pruning and reduction system. It's not a bug, it's a feature ...

For what is worth, I've seen how Rybka and Naum behave on these situation. This is an example:

Depth 19 - 19.d4 is best move with score 0.60
Depth 20 starts - 19.d4 fails low as 0.45--
Engine starts analyzing the alternative moves
19.Nf3 is discarded
19.c4 is discarded
19.e4 fails high as 0.75 ...

And what should we do when facing fail-low + fail-high combo?

What about resolving the real score of the move?

There is no such a thing as "real score of the move". Just pick up any very strong engine and a complex position, turn on SMP mode and do a fixed depth search many times. You get a ...

hyatt wrote:
Namely that on a fail-high followed by a fail-low, you believe the fail low is correct every time. Bogus concept.

And you are suggesting that we should believe fail-high instead... I wonder what makes that less bogus.

And what should we do when facing fail-low + fail-high combo?

How can it be correct to get a fail high and then not play that move??? If you fail high and it is not a better move, that's a bug.

Haven't you ever heard of search instability???

This is exactly what happens when you combine a) minimalistic aspiration window b) aggressive late move pruning ...

Great work!

I can see that you've learned many things from stronger engines, but I think it's only natural way of learning.

Some points behind 2..d5!? (It is the reason why I don't want to play 2. f4 as white :D).

White must react:

a) 3. exd5 Nf6 is a sort of scandinavian where white has extra move of f4 and black has
extra move g6. I'd say g6 is much more logical move in that opening than f4, so black should be okay ...

Hagen wrote:I'm having trouble as Black countering this move order my opponent as White makes and I'm wondering if anyone can tell me a specific series of moves to counter this. Thanks.

One option is 2..d5!?

for KNKP my recognizer returns drawscore as an upperbound for white. The search than proceeds as normal.

actually this is a good illustration of dangers inherent to coding interior node recognizers and forgetting about exceptions. There is a KNKP position with a rook pawn, the king of the weaker ...