Disparity between batch simulation and BIB

As per: https://twitter.com/AskMrRobot/status/930286917433921536

New users can only post two links, so here’s a pastebin with my export string and relevant links for this post: https://pastebin.com/wHEXLARS

So the long and short of it is that I’ve got two conflicting data. I’ve run a batch of simulations comparing various combinations of legendary ring/other ring/robe (several times, in fact) and received a recommendation as to which setup sims best on Goroth using my APL.

I’ve also created a ML strategy based off that APL (several times, in fact) and received a wholly different recommendation from Best In Bags, seemingly contrary to the actual results of simulation. The recommendation that I receive from BIB, Fetid Gutcover Apron/Ring of Contempt/Soul of the Archmage, consistently sims lower than the best result of the batch simulation, Robe of Aqueous Command/Seal of the Second Duumvirate/Soul of the Archmage.

I’m not particularly sure what’s going on here, and would enjoy some clarity on the subject.

The short answer is that the results of those sims are very close together. The one recommended by BiB with your custom gearing strategy simulates to 0.3% less DPS than the highest combo that you tested. That is a very close result, and is considered well within the predictive powers of the gearing strategy.

When you create a gearing strategy, it builds a predictive model that will calculate the DPS of any given set of gear without actually simulating it. It won’t match simulations “exactly” - but it will be able to get you so close that there is no practical difference in-game - and it will be able to look at every combination of gear in your bags (of which there are millions or billions) instead of just examining a handful.

What gearing strategies + best in bags does is: test every single combo of gear you have and choose one of the sets of gear that simulates within 1-2% of the theoretical best. A custom gearing strategy will only be valid for the legendary/set items you created it with, just FYI - the error margin goes up if you change those legendary/set items.

In order to simulate even a small fraction of the gear combos in your bag would take prohibitively long. In your case, you decided to simulate a handful of item combos, so you can just lock in the items that simulated highest if you prefer. With differences that small, though, it doesn’t really matter which combo you use in-game.

We’re slowly trying to convince everyone that simulators are really blunter instruments than people have come to believe that they are, but it’s an uphill battle :wink:

Well, I may be an edge case: I naturally gravitate towards trying to make lower haste solutions work, as I have a bit of an issue with my wrist that flares up past a certain amount of APM, so I’m constantly picking and poking at gearing strategies to see if I can find gearing foibles I can exploit to justify run with less haste.

What I’m getting from this is that I should use the gearing strategy to do 99% of the work and then drill in that last mile with some targeted simulations, as I’ve essentially done in this situation. Which sounds perfectly viable, frankly.