Yes I see. The paper talks about what to do when null move fails-high while here we are talking about what to do when null move fails low, so I think it is perfectly orthogonal to what proposed here.hgm wrote:Note that the verification Ron is referring to is exactly the complementary case of what you worry about, namely not taking a null-move fail high for granted, (taking a beta cutoff), but doing a reduced search if it occurs, and use the result of that.
This is a good point. This horizon effect seems to be the reason the tests do not show this prune technique is clearly better...but hold, we have more to say about thishgm wrote: the veriification search will show you if the retaliation works in reality, as the opponent will start to work harder on the original threat to make up for the loss in the SEE. The main problem I perceive is if you can afford reduction of the verification search. My guess would be not, as retaliation against a target unrelated to the true threat is the most common source of horizon effect, especially if it is a valuable target: the opponent is then forced to recapture first, and this might push the original threat move over the horizon.
Please read the next post. It is also better explained the SEE stuff.
I prefer to do a different post because after some night thoughts (and some hundreds games of testing) I have finally understood what we are talking about and so I have also changed the name of the technique in "Null driven IID"
Marco