One question about stability of a simulation

Hi !

I have one question about the simulator. “Error target” is like “how many time this sim will run” (this is a limit during a loop in the code i guess).
Std Deviation is the variation for the value (in it’s stasitical meanning)?

Like for this one : https://www.askmrrobot.com/wow/simulator/report/689e543598a6420eb910072e3434c07a
I have setup the Error target to 0.5%.

What i understand is :
This simulation have run till the variation of something got under 0.5%
But the result of this sumulation can vary by 2333 around the average.

The value showed is the average ?
If so how can i make sure the simulation is stable (the one i shared using default setting keep vary by 2k HPS around the value some for death chance).
How can i stabilized it ?

Yes – a single “simulation” is actually a bunch of “iterations” of the same fight, then averaged together.

The big HPS number shown at the top is the average HPS of all of the iterations that were run.

The error target controls how many iterations to run, and is directly related to variance (standard deviation) – you can calculate the margin of error from the variance and the number of iterations run. We keep going until that margin of error drops below your specified threshold.

(Calculating margin of error has some nuance to it, but we use a very standard approach – a 95% confidence interval.)

We would have to modify the code to allow using ally deaths or something else as the metric that determines how many iterations to run. Right now we only use DPS, HPS, and TUF. We use these metrics because they tend to be both useful and less volatile than other metrics. e.g. for a tank, if you use straight-up death chance as a metric it has a much larger variance than TUF, so the simulations take a lot longer to run. (That is also one of the several motivations behind creating TUF in the first place – something that is useful and meaningful but also less volatile.)

I’d have to look into whether we could allow other metrics to control the length of the simulation or not… parts of the code may rely on the current metrics to work correctly right now.

Having it stabilized on the HPS does seems to be the better idea, “allies death” is quite volitile.

So, i’ve put the Error target to 0.05% instead of 0.5% the sims took way longer but now the HPS isn’t moving that mutch, so i will probably do something like this for the “test” i’m doing with fistweaving.

Thanks for your time ! =D