What you are asking for is the uncertainty of the gage.
The statement "My rounout gage measures to +/- X microns." You need to know what "X" is, then you can emplace a guard band the size of X and run to tighter limits than the print specification, running limits. This guarantees your parts are in specification.
Sadly, this uncertainty number is not usually calculated in an MSA/Gage R&R sheet, but the calculation is pretty straightforward.
The first way people do this is to take the 95% confidence on the variability of the gage, which is 2 standard deviations. In a typical Gage R&R study, there is a column of standard deviations, usually marked "SD". (There is ALSO a column typically marked (6*SD), you do NOT want this column. You also do not want the percentage columns. You want the first column, again, marked "SD". The row marked "Total Gage R&R" is the one you want. Double this value (2 standard deviations) and that's your 95% confidence uncertainty. So lets say you had an SD of 0.0007 for Total Gage R&R, then your uncertainty is +/- 0.0014. If your print spec is .05, you reduce this by one side of the uncertainty and run to 0.0486. (I'd probably round that DOWN to 0.048).
The second way people do this is incorporates NDC analysis. Again, you need the SD column, but this time you want the "Total Variation" row, not the "Total Gage R&R" row. You take the Total Variation number (TV) and use this equation:
Uncertainty = 3*TV/NDC
where NDC is your Number of Distinct Categories. These numbers are usually very similar. I calculate both and take the worst of the two, just to be safe.
BAM. There's your guard band width.
(To be complete, if you are using the old Xbar-R method, I have seen it said you can get uncertainty from an intermediate value using this method. But I personally don't like Xbar-R at all. It was good back in the day before computers because it estimated sigma and used lookup tables for shortcuts. But with computers today, it's silly to take a calculation shortcut. A lot of these statistical methods are from the slide rule and earlier era, where there was motivation to come up with "approximate" methods to save calculation time. That's not the case today .... In any case, if you hunt around on an Xbar-R sheet, you will find an intermediate calc called UCL_R. It's found by looking up the average range in one of the tables. You take this number and divide it by 2 and that's sometimes reported for uncertainty. Again, it follows the upper two methods closely. I don't like it, but I wanted to be complete.)
And that's all the ways I've seen people do this calculation.