I'm not quite sure how to define observation with no circular aspect to its definition. I mean any type of instrument (including a camera, human eye+brain, a dial on a meter, etc) capturing a snapshot of information. The description of a device includes its state of motion or rotation.
Within a particular theory, when physically interpreting its mathematical constructs, I require that internal conventional changes do not affect what I match to physical measurement. What these features are, depends on the theory. In GR, it means diffeomorphisms. The subtlety, is that a mapping that changes, say, a value of (x1 -x0) from 1 to 3 will correspond with a change in metric that makes the length stay the same. And I don't consider units of measure part of this at all. Using the same example as chirality, I get to units as saying I have a reference object for meter, for foot, etc. Under diffeomorphism, any measurement referenced to these objects stays the same.
Finally, I note, that within GR, to have the right properties (as above) a computation deemed represent a measurement must be a collection of scalars. Really, this could be viewed as a collection of individual measurements. Thus, a color photo of something is really a measurement for each pixel on a CCD or grain on chemical film.