I played around with this a little bit… it’s kind of a tough case.
Firstly, with your first snapshot, you’re right on the edge of where we generated simulation data for our statistical models (around ilvl435 equipped is the minimum). Once you get outside that range, we have to use some different scoring methods. We have code to make that transition “smooth”, but in this case there was a little bit of a discontinuity. Our next update will add some code to smooth that out.
That aside, once you start using custom azerite rules… it gets much, much harder to guarantee that it will find the highest scoring solution. It will always get close… but might not always find the best. There is no way around that unfortunately… the optimizer would become painfully slow if we tried to give you a mathematical guarantee that no other set of gear could possibly score higher.
I’ll see if we can tweak it for this case a bit… but you end up in a “wack a mole” kind of problem: a tweak to the algorithm that improves this case might not work as well in other cases.
For now, change your azerite rule to only exclude dagger in the back. You get a better result than the one you found in your second case, and it has one deafening crash as well. In general: use as few and as simple azerite rules as you can to get your desired result. Exclusion rules are the simplest and will have the least impact on the optimizer’s accuracy and speed.