hgm wrote: ↑Fri Aug 05, 2022 9:03 pm
What you describe is not at all the same as what I described. You order all moves by SEE (key = SEE). Consensus for what is best is key = (value[victim] < value[attacker] ? SEE : value[victim] - value[attacker]/16.).
Aha, I see. Probably I missed this because I didn't read the whole thread.
If this works for you it doesn't automatically mean this would be best for somebody else too. Don't you think that I tried many things in the 45 years that I'm busy with this stuff? Sorting moves in quiescence by plain SEE scores always gave me the best result, but your mileage may vary.
I did not say it worked for me; I said the concensus was that this is the best ordering. This topic has been discussed many times over the years, e.g. by Tord Romstad and Bob Hyatt, and after extensive testing by the involved everyone always seemed to agree that this ordering worked best. And on theoretical grounds it is expected to be best for independent exchanges. The probability that exchanges are dependent is not very large.
hgm wrote: ↑Sat Aug 06, 2022 8:40 am
I did not say it worked for me; I said the concensus was that this is the best ordering. This topic has been discussed many times over the years, e.g. by Tord Romstad and Bob Hyatt, and after extensive testing by the involved everyone always seemed to agree that this ordering worked best. And on theoretical grounds it is expected to be best for independent exchanges. The probability that exchanges are dependent is not very large.
Maybe I'm stubborn, but I don't care much about 20 years old 'consensus'.
Well, don't kill the messenger. I can only point out that there are many positions where you have to choose which one of multiple captures you want to try first where pure MVV/LVA or pure SEE ordering will obviously pick the wrong capture, and that the algorithm that most people have been considering best for 20 years in general gets these right.