Trouble understanding the alignment of epipoles with an axe

Question

This question arose when I was reading this paper. The goal is the estimation of the relative pose of two cameras C₁ and C₂, given five point correspondences.

The projective planes are considered as unit spheres, and the global coordinate system is chosen such that the Z_G axis joins the camera centers. The problem is solved if we can find two matrices R₁ and R₂, respectively mapping the internal coordinate systems of C₁ and C₂ to the global system described previously (to me, it sounds that these matrices will just align the Z_{c_1} and Z_{c_2} axes of the cameras with the global Z_G axis).

These matrices are computed in an iterative manner, and at each iteration, all point correspondences (v₁,v₂) are updated via a rotation, becoming (R₁v₁,R₂v₂).

So far, everything makes sense. But later, the authors state that these rotations change the direction of the epipoles (I assume relative to the camera Z axes), and that "by rotating the unit spheres such that the z-axis Z_G is aligned with the epipoles, i.e. Z_G=R₁e₁=R₂e₂ the rotated point correspondences (v₁,v₂) become coplanar with the epipoles e₁, e₂".

To me, the Z_G axis, being the baseline, is always aligned with the epipoles, and a rotated epipole Re is no longer an epipole, since it moves away from the baseline. But the quotation above implies that there are configurations in which the epipoles are not aligned... As you can see, I'm completely confused... please help me understand what the author means by that.