Saving ifile database source files

I have a preliminary set of ksh93 scripts which can process Maildirs and
update ifile databases.  As part of the interface, I plan to keep
"shadow" Maildirs corresponding to the last two weeks worth of mail in
each ifile category.  So, if you have the normal $HOME/Maildir to
receive mail, I plan to also create $HOME/ifile/Maildir,
$HOME/ifile/Spam, and so on as Maildirs for each ifile category.

My question revolves around keeping the source files for the ifile
classification database.  Mozilla uses a Bayesian filter and only keeps
the Junk mail for two weeks by default.  Much of the discussion on
setting up ifile involves downloading and organizing multi-megabytes of
database source files.  I do not understand this.  For one thing, it
involves trusting someone else's idea of spam and for another, it
involves a lot of disk space.

Currently, I plan to delete old database files to keep the directory
sizes under control.  I do not plan to update the ifile database when I
delete the old files--I think it should just continue to collect words. 
If something happens, I can use the two-week archive to re-seed the
database.

Does this sound reasonable?  Why spend so much time on the website
tallking about organizing huge quantities of mail if all you only really
need the word counts?

/Joe