Ok. Lets imagine you have 50mm prime lens.
And try to imagine that your lens is not perfect. And have some distortion in top left corner. But this distortion still can be estimated with Brown or Brown+tangent lens model.
So all images in your dataset have exactly the same lens distortion model.
But camera add rotation tag depend on how you rotate camera.
And main problem that this tag can be not correct.
Or Photoshop will unrotate image but forgot delete tag, Or rotate your image to -90degree and delete tag and other to +90 degree and also remove tag…
So we can’t trust this tag so much.
But if you remove this tag from your RAWs right after copying from SD card. You will have all images horisontal with top left corner exactly the same on all images.
And main advantage, that all this images can be grouped in one lens group. So undistortion model will be estimated from all images and features detected on them. This mean more precise results.
But as soon as focus can give some small differences, before final refine Align better ungroup images and RC will finally adjust all undistrotion models per image.