This is a test document used with the API test.  This paragraph must be at
least three lines (including the blank line) to be counted as a "paragraph".

This file will be indexed by paragraph, and the simple query will search
for the word "word".  We'll expect the mset to contain two documents, including
this paragraph and the fourth, below.  Since this paragraph uses the word
"word" so much, this should be the first one in the match set.  I'll just say
the word a few more times (word!) to make sure of that.  If this doesn't word
(typo, I meant work), then there may be four-letter words spoken.

This paragraph only has a load of absolute rubbish, and nothing of any use
whatsoever.

This is the other paragraph with the word in the simple query in it.  For
simplicity, all paragraphs are at least two lines, due to how the hacked up
indexer works.

This is another paragraph which won't be returned.  Well, not with the
simple query, anyway.

And yet another.  This one does mention banana splits, though, so can't be
that bad.
