corruption

You may have noticed my website is down.  pyds-start stopped working and
i began investigating why.  Apparently TONS of my .data files have been
SO corrupted that gedit/gvim/etc says they are binary files.  ARGH.
Maybe they are, but they are still causing strange non-ascii chars next
to url's in my html and xml files in my weblog and aggregator.  I've got
the html and xml files stripped of the badness BUT the .data files are
still corrupted.  It seems like the ML files are being created out of
the .data files.  Is this true?  Is there some way I can "reverse" the
data->ML process to get a clean set of data files?

-- 
Jeff