Nope, I was wrong, "score == alpha" is a fail low of course - in fail-soft and fail-hard. But we don't need to re-search if we set root-alpha at the start of a new iteration to an exact or lower bound mate score found in previous iteration - if we trust that score.John Major wrote:Or,Gerd Isenberg wrote: You should use < alpha, == alpha is not a fail low.
< alpha when using fail-soft, <= alpha when using fail-hard
That has nothing to do with fail-hard or soft. In fail-hard, "score < alpha" does not occur, but "score == alpha" has the same meaning of an upper bound - score might be less.