The two model approach will not provide both sets of "stress due to sustained loads" for a proper calculation of the allowed expansion stress range according to (1b) [what we have called the liberal allowable].
If you create a separate model holding only those restraints that are active in the operating case, the calculated expansion stress range from that model will not necessarily reflect the "installed" response influenced by those now-missing supports.
You should take a look at CAESAR II 2016 where we introduced the Alt-SUS load case. We run the standard cases you have come to expect but we added a new load case where the sustained load vector is pushed through the same (converged) stiffness matrix that produced the operating response.