Three things nail this down:
1) Set the view origin to something in the model that cannot change - such as an initial Coordinate system that precedes all other geometry.
2) Make the view a partial view - this fixes the view boundary so that notes and other items won't drift off. Just surround the model as-is, preferably bigger than minor changes will exceed.
3) Additional option: Create snap lines offset from geometry (not view boundary)
View origin is essentially a push-pin/thumbtack to nail the model to the paper. Leaving it set to center keeps the center of the geometry nailed down, but where that center is changes as the model changes and with it, the apparent location of the model on the paper.
I always wondered if the developers were willfully ignorant or thought they were doing a favor by driving 2D notation location relative to the view boundary. Of all things, nothing is more wastefully time consuming than trying to fix a drawing** when a minor change has been made and all the notations move. If I was prompted to "pick a location at a random fraction of some dimension of the view extents between the view origin and the point where you want the geometry to appear, but will later be moved upon re-scaling of the view extents" that would be a heads-up.
When I make a mark on something I expect the mark to be in exactly the same place later.
Instead the PTC developers for the drawing/detailing package seem to excel at creating dependency chains to do exactly what no one would seem to want to happen - moving markings on drawings.
This would make an excellent addition to a PTC Blog "Why we move the draft entities because the model extents change."
**or taking the steps required to avoid having to fix the drawing
