Custom Gearing and Relics


I’ve been toying around with an alternative build for Moonkin based around the new legendary ring (soul of the arch druid) and Stellar Flare. The goal of it is to keep Starfall down and only use starsurge as a filler to avoid capping AP. The tweaks to the default APL were pretty simple and the results have been good (you only lose about 6% dps in pure single target while obviously gaining quite a bit of free cleave). The idea of it is as a go to build for fights with lots of cleave but important single target damage.

Anyway, I ran a custom gearing strategy:

The results, unsurprisingly, came out favoring Mastery a bit more than the default AMR single target machine learning weights.

When I went to calculate BiS using this custom gearing strategy, however, it still equipped the same relics as when using the Default machine learning gearing strategy. Most problematically, it equipped a Starsurge Relic when it could’ve equipped a Falling Star (Stellar Empowerments) relic of the same ilvl. Manually swapping to a Falling Star relic improved DPS by over the margin of error (Falling Star relic, Starsurge relic)

Is there any way to build in relic traits to their value via custom gearing strategies? It seems like they’re just stuck with the default gearing strategy values. Is that right?


Artifact Traits are calculated and saved with the gearing strategies that you run.

Those two sets of gear are within 0.65% of each other. That is well within the target margin of error for Best in Bags, which is 1%.

Each individual simulation can give you an error in the 0.25% range, but the predictive model has to take into account stats, legendary items, set bonuses, trinkets, relics, etc. Getting within 1% is excellent and definitely working as intended.

You can certainly do your own spot-checks to squeeze out that last theoretical 0.5% or so DPS, but the instant BiB optimization can’t get that perfect.


I see, thanks. Yeah, when I said outside the margin of error I was referring to the sim margin of error but not the predictive. I didn’t mean to imply that within 1% wasn’t good enough, either. I just was going by the wrong margin of error.

Cheers and keep up all the awesome work!