assumptions of anti-corruption layer approach?

hi,

i can see how the idea of A.L. might work when a new feature is very
on-the-side compared to old crufty code.

i don't grok how one can add a new 'feature' into an old crufty system
via A.L. when the feature is a refinement of some part of the internal
state of the old code. does one break the old code into 2, before vs.
after the 'new feature', and insert the new feature in-between? or
something?

thanks for any musings.

------------------------------------