In your scenario, you don't know whether it is severe or not, so you just want to provisionally mark it as severe, and have someone with the appropriate domain knowledge make the final call. Until that happens, decisions made on the basis of that provisional call could be bad decisions.
And what might those decisions be? Priority, for one! To say that the severity should stay at 'severe', and the adjustment should be made only in the priority, is to confuse the two distinct concepts. I see a lot of that going on wherever data are used.
If you want to characterize bugs by the sort of technical outcome they produce, then do just that, instead of trying to bend 'severity' to try and encompass it, while actually hiding the issue that you wanted to characterize on. Again, while technical outcome is an aspect of severity, they are not the same thing (and, FWIW, crashing is rarely the worst thing that could happen (unless the domain is vehicle control.))