What is the difference between RMS of distance to ray and RMS of 3D errors in context capture ?

And which one correctly leave the positionning information of the model ?

I don't get the difference between them.

Best Regard.