Set bonuses are difficult to deal with sometimes, so the optimizer can get stuck and not check a few potential setups that would be better. We try our best to make the optimizer always get within 1% of the absolute best setup that could potentially be found. Considering how many setups of gear it has to go through, that is what we have found is the best trade-off between speed and accuracy. Bute-forcing it to always, always get the perfect answer would take a very long time to calculate.
We can plug this example into the our development version and see if there is a way to improve it without slowing things down too much.
I’ll try messing around with it a bit… the results are pretty close either way though, still well within a range that would be near impossible to discern in-game.
The problems with optimization algorithms is that improving one case can often make another worse, so it takes a little while to test out alternative techniques.
It’s all good. I can tinker with it on my own. It is actually good to know there is 1% margin.
On another note. this might deviate from the thread, but do you guys account for Diminishing Return on Haste? The reason I’m asking is that if I choose to use Haste Pot, AMR drops my Abacus (AP + Haste on Use) and chooses my Bladefist Breadth in its place (Crit & AP on Use).