In case anyone is intrested, I received a reply from Robert Brown:
Here is his answer:
QUESTION: "I'm in the process of developing some type of minimum TAR for Micro Balances. Since the best weight will not guarantee a 4:1 or even a 3:1 ratio, I need to come up with an acceptable level TUR value for these balances."
TUR is an issue when testing to see if the measured value complies with specifications (a failure could result in an "Out of Tolerance" report).
CASE 1: No Specification Compliance:
If there are no specs for the weights and/or...
1) The weights are only used within your quality system
2) Correction factors are on the calibration report and applied when the wieghts are used.
3) No one depends upon the weights being within a secification of nominal.
Then Out of Tolerance issues do NOT apply. Be sure that the UNCERTAINTY of the Balance measurement (is calculated by squaring the repeatability of the balance, and squaring the uncertainty of the weight we use to check the balance) is DECLARED, and USED when the weight is used to make measurements with declared uncertainties.
CASE 2: Specification of Compliance:
If on the other hand you are verifying that a weight has not drifted too far from a nominal value, and that would result in an out of tolerance condition, then TUR applies. The key problem is that the user of your weight will not be applying correction factors and will be depending upon your assurance and the width of the specification.
In this case,
ISO-17025 p18.104.22.168 "The calibration certificate shall relate only to quantities and the results of functional tests. If a statement of compliance with a specification is made, this shall identify which clauses of the specification are met or not met."
This requirement of ISO-17025 is often interpreted as "guard-banding". In a recent (Aug 17, 2003) paper by David Deaver, "Good, Bad or Indeterminate" there is the only statement that I have seen that refers to least acceptable TUR = 1.5, BE CAREFUL! If you allow a TUR of 1.5 then guard banding will KILL you!! Many false failures will be the result. Dave's recommendation is part of a larger guardbanding strategy (that I support)
1) No guardband if TUR > 4
2) Guard band of 20% for all 1.5 < TUR < 4
3) Calibration NOT ACCEPTABLE if TUR < 1.5
A test limit that is tighter than the actual specification, in order to avoid false pass errors.
I also had a good conversation with Robert on the phone (he called me) to get a better clarification of the subject matter. Nice to see people like Robert helping out others in the field of Metrology. Robert L. Brown was presented at the NCSL International 2003 Symposium GUM Uncertainty Analysis With No Greek Letters
In there is a link to his PDF file.