Hello!
I like notmuch a lot, so I'm writing a (conceptually) similar software
about addressbook: it will scan all your emails, storing email addresses
in a xapian database (you can think of it as little brother database[1] on
steroids)
The part that I'd like to re-implement is "notmuch new": it seems that
in the xapian db there is not only informations about each mail, but
also the mtime of each directory. My impression is this being "chaotic",
but probably I am just missing the point.

So, here's the question: how is the db "structured"? is there any
documentation to look at?

[1] http://www.spinnaker.de/lbdb/