Sorry I missed your call. Please leave your message at the beep.
nak81783 pretty much pointed it out though: FPS variation will vary based on paint. In fact, I believe it actually dominates the shape of your distribution curve.
For instance, I ran the same batch of paint through the same barrel, same day, but on 2 different markers here:
Look at 2013-06-29, with Automag and Spyder. X is FPS, Y is frequency.
The center of the curves (the velocity setpoint) were different, but the shape of both curves appear very, similar. Standard deviation was 6.17 vs 6.19.
In that example, I was essentially holding the paint constant and varying the gun.
Doing so in a repeatable, verifiable way (such that many people can reproduce and validate the results) is tricky. You only have so many shots before you have to open another bag, and who knows what you'll be getting with that second bag.
You really need a reference sphere that everyone can obtain and agree on. And you need enough of them. And a reference barrel to go with that sphere.
I think this came up in the CCM SR1 development as well. Read the description in the video -- "it ain't the gun".
So you either have the reference sphere and reference barrel (which themselves require validation), or you find some way to measure and normalize results based on deviations from the reference. Uh... good luck with that, I guess?
But the sad fact is most guns in a "modern" (oh god I'm starting to use that word now) configuration can easily do this task pretty well. Well enough that the error from the gun is probably dwarfed by by the error from just the paint/barrel. Paint nowadays is way better than it used to be, but it is still (measurably) imperfect.
"Accuracy by aiming."
Definitely not on the A-Team.