Are you using one of our default gearing strategies? Sometimes discrepancies like this can pop up because a single default strategy will never be “perfect” for everyone.
We are working on making our defaults more adaptive, but in the meantime, you could try the beta of our new custom gearing strategy technique, see if it gives you a better result.
To do that, on the simulation page, choose “BETA! Strategy V2” from the “Sim Type” picker. You will need our client program to run it, since they can take a while depending on your computer’s speed. (If you have a relatively new quad core and run in max performance mode, the new strategies should take around 30 minutes.) When it is done, check the Machine Learning option, then press “Save and Use” on the report it produces, give it a friendly name. Refresh the gear optimizer page, and choose it from the strategy picker at top/middle of the page, then do best-in-bags again.
Let me know how that turns out! This is a pretty early-access beta, but in theory the new system should handle anomalies like this much better.