[racket-dev] Confusion of difference of evaluating syntax vs compiled code?

I'm running into dynamic evaluation behavior that I don't quite
understand yet.  My example is:

    https://gist.github.com/dyoo/5314045

It's meant as an experiment to see whether it's possible to avoid 3d
syntax in certain places like the gui-debugger.  I try to throw in a
wrench on lines 52-54:

    https://gist.github.com/dyoo/5314045#file-annotation-example-rkt-L52-L54

but thankfully, when I evaluate this in DrRacket, I get the results I expect.


However, that's when I'm evaluated the annotated+compiled code.  If I
modify the example so I'm evaluating just the annotated code, by
modifying lines 117-119
https://gist.github.com/dyoo/5314045#file-annotation-example-rkt-L117-L119

from:

    ;; But we can evaluate it in ns:
    (printf "evaluating the annotated code\n")
    (eval compiled-code ns)

to:

    ;; But we can evaluate it in ns:
    (printf "evaluating the annotated code\n")
    (eval annotated-code ns)


I get the unexpected shadowing I'm trying to avoid.  Why am I getting
the clash here, and not in the compiled code?