I see, maybe I should cut back on my requirements :-)
The "documentation" part applies to every kind of software, even to R skripts. The overall goal should be that the author himself is able to understand what he did 4 weeks after he last looked at his creation.
As with the "paradigms" of software engineering, I was thinking about e.g.
a) information hiding in the sense of component orientation (use namespaces, don't use global variables etc),
b) separation of concerns (don't mix GUI and model code),
c) use unit tests
etc. which all kind of don't apply to short R scripts.