My preference is to do an attribute gage r&r on testers. Their goal in life is simply to determine if a part is good or bad, and provide virtually no information on process adjustment or control. Variable gage R&R, therefore, is overkill. Therefore a 50 pc study replicated over 3 operators (or 9 replicates if only one operator with no operator influence on results from loading and unloading - using "time" an "operator" using time gives you stability, too) and do kappa study. 25 pcs targeted to have detectable defects that the machine looks for as a part of the 50 pc. study.