> Comment are interpreted by computers. At least they are parsed. :)

Yes, the form is interpreted but the content is not :)

> Metadata wouldn't be metadata if they appeared in the actual data, so
> they have to appear out of band.

That is true in the general case.

However, there's a minor pitfall in assuming that there can be a clean
separation between data that cannot be ignored, and metadata that can.
In practice, things that are added as advisory metadata may later turn
out to be mandatory information and things will break if it is ignored.

For practical applications, it may be that the data/metadata distinction
is most fruitfully explored from the standpoint of whichever science/art
deals with things like schema evolution and information classification.
Something like library studies or information studies? In any case, it's
going to be some messy "soft-science" stuff that we don't necessarily
enjoy around here but often have to live with anyway :)

That's why I would advocate having just one layer of data and coining
new macros. The macros can be no-ops at first so the data is ignored. If
it turns out later that parts of it shouldn't be ignored by Scheme, then
the macro definition can adapt.