> However, it still fails validation at
> http://feeds.archive.org/validator/

The version this morning seems OK.  I suggest you keep an archive of
these bad feeds under static URLs, so that we can more easily see the
problems.
 
> The problem is "bad characters". I'm not sure how to start to debug
> that. What is the first thing I should look for?

Most likely thing is the use of HTML entities (like é ) that
aren't valid because RSS is an XML protocol and doesn't recognise
these HTML-defined entities. A _very_ common error in RSS feeds.


> I wrote the 0.91 script (a PHP script) first because I assumed it
> would be easy. 

I have to ask why anyone needs to write RSS scripts these days ? 
(Although I spent yesterday doing it myself).  There are very many
already out there, and it's a rare situation that really needs
something written from scratch.

I suggest that you learn a bit more detailed XML (entities for one
thing, namespaces for another) and learn how to read a formal DTD.
It's one thing to make a feed work once during testing, but quite
another to make a reliable feed that handles all the data it will meet
over its lifetime. Sadly RSS tools suffer badly from this - they run
for a week, then crash when they meet their first accented European
character. To get a reliable feed, you really do need to know how to
_understand_ the specification, not just match up one example.

As to the versions, then I'd support 0.92 and 1.0   1.0 is best, 0.92
solves compatibility issues for some older or simpler aggregators. 2.0
is pointless.

You don't need to understand Dublin Core to use RSS 1.0, but you will
do if you want to really use it. Worth the effort of studying it
anyway (Learning Dublin Core is vastly more useful than learning RSS
anyway)

Hang in there - RSS versions don't really vary that much between
themselves.