Difference between `Rigid.make_transform_from_reference()` and `Rigid.from_3_points()`

What is the difference between `Rigid.make_transform_from_reference()` and `Rigid.from_3_points()`?

They are not aliases of each other, since on random input coordinates they give different outputs.

If I had some backbone atoms and wished to make rigid frames, which operation should I use and why?