Epistemic Assumptions: Are They Assumed to Be Backwards Vindicated or Forwards Vindicable?

In a paper from 1997, entitled Implicit epistemic aspects of constructive logic, Göran Sundholm introduced the term epistemic assumption for the assumptions with which the semantic (contentual) justification of a formal rule of inference invariably begins. The question to be discussed is whether these epistemic assumptions are assumed to be demonstrated or merely to be correctly (properly) asserted in the semantic (contentual) sense. I shall favour the second alternative.