> I feel like it is not a huge restriction for practical applications,
> but I may be wrong. I would say that keys should be stored as is, but
> compared as strings.

Thinking about it some more, I am wondering: is it more important
- to distinguish keys which only differ by their catcode, or
- to allow for any token as a key (e.g. Hef{}feron, which currently
breaks the delimited argument approach) ?