I can shed a bit of light on this. The constraint solver in any CAD system works pretty much the same. It just collects a set of geometries, constraints, and dimensions, and tries to solve them. It does not 'know' that this is a center-point rectangle. What it sees for this is:

- 6 lines
- 5 points
- 2 dimensions
- 6 geometric constraints (parallel, perpendicular, horizontal)
- 10 coincident constraints (which bind the points to the lines)
and, the solver just tries to solve all of that geometry and constraints. There is no such thing as a "center point rectangle" to the solver. You can create the exact same system using the line command and the constraint/dimension commands. If you adjust one of the dimensions, there are many solutions of this under-specified set of geometries and constraints, all of which are valid. Technically, there are an infinite set of solutions, but, practically, that is not really the case. The point that everyone here is making is: Unless you give the solver a bit more of a hint about how you would like this system to react, it can choose which ever solution it "wants". Most likely, it will choose the solution that converges the fastest. I guess that means moving one of the points in the rectangle is the default solution.
Jeff Strater
Engineering Director