Well, reducing less is much more expensive then just sorting it earlier. If the move helps, it should already be apparent at QS level. A problem might be that interpositions are inherently risky, as the square you move to is known to be under attack, and you have no guarantee it will be protected. (In fact by moving there you just decreased the number of protectors by one, if it was not a Pawn move.) So the chances that the piece is simply captured to renew the threat is appreciable. And even if it succeeds, the interposed piece will now be soft-pinned, which creates a weakness.cetormenter wrote:IIRC Stockfish had a concept similar what you are describing here. Whenever a null move failed low, SF looked at the move that failed high for the opponent. Every move that prevent this move from being able to be played (by moving in between the from and to squares) was then reduced less than it normally would be (it may have also removed these moves from being LMP candidates). However this all was removed somewhere around SF5 or 6.
I played with a similar concept inside of Nirvana but I could never get it to perform even neutrally. I suspect that this is due to the fact that multiple moves can cause a null move to fail low. In the case where there is more than one move, we are just randomly choosing moves that we are going to reduce less.
Withdrawing the threatened piece is in general a better solution. Unless the piece itself was soft-pinned. For weak pieces, with few moves, (safe) withdrawal is often not possible, but such pieces are easily protected. Unfortunately it is quite hard to selectively geenrate moves that protect a given square. Anyway, these type of decision are a next step; first the threat has to be identified.
And that is another weakness of the Stockfish method you describe: the move that refutes the null move is very often (perhaps usually) not the threat, but just an unrelated trade of a higher piece, so that preventing it buys you absolutely nothing. This is the problem that has to be solved first, before it can be judged whether any particular counter measure to the threat is beneficial.
If there would be multiple independent refutations to the null move, there is nothing you can do anyway. Your real moves will also fail low, because the opponent will cash at least one of the threats. You cannot do something smart here, as you won't be made aware of the second threat. But if you fail low, it also doesn't matter much how you sort the moves. (It would hurt to reduce less, though.)