From Mark Foster ---Applied Geometrics---(different thread, but I think applicable on what we are trying to talk about here---order of clamping)
IMHO first 2-3 sentences are applicable here.
Begining of Mark's post:
Your "Go" gage should FIRST, simulate the datum reference frame -- that is, it should establish the part in its correct (specified) coordinate system; and then SECOND it should see if there is no material where there is not supposed to be material (i.e. the "go" gage should "go").
Then the "No-Go" gage should FIRST, simulate the datum reference frame, and SECOND it should see if there is material where there is supposed to be material (i.e. the "no-go" gage should "not go").
Your gage elements for simulating your datum features on the part should be the virtual condition sizes of the datum features, and must always go, and then by the amount that the actual datum features have departed from those virtual condition boundaries, you are allowed to "wiggle the part" (i.e. take advantage of the datum shift permitted by the MMB modifiers on the datum reference frames of the profile callouts) until you can make the "go" gage "go."
And you should never be able to make the "no-go" gage "go" no matter how much wiggling you do of the datum reference frame.
End of Mark's post