Imagine two user stories (=> Backlog Items) A and B. Both stories share some complex common low level logic. This logic is not yet implemented. Let's say that it takes 13 story points to implement it. The rest of stories A and B takes 3 story points each.
How would you estimate such stories during backlog refinement if they aren't close to implementation? Because depending on which story is gonna be implemented first, the stories will take either 3 or 16 story points, which is quite a difference.