[Cc'd to Peter Mutsaers, whose name is in the source code as the
author of the orphan scoring functions.]

I found two independent problems with orphan scoring in pgnus 0.97.1.
The first is that articles which have no references have this indicated 
internally in gnus sometimes with "" and sometimes with nil.  The code
only checked for "", and so many articles that were actually roots 
were thought to be orphans (along with all of their descendents).
The fix for this is a simple change to one line.  I don't know where
the discrepancy between "" and nil originates, but it might be worth
tracking this down.

The second problem I noticed with orphan scoring is the following.
Suppose we have three articles, A, B and C, with B a followup to
A and C a followup to B.  If C has no references header, but does
have valid In-Reply-To header pointing to B, the current code will
conclude that it is an orphan, because the current code tries to
find the message-id of a root article in the headers of C.  

To fix the second problem, I completely rewrote the orphan scoring
code.  It turns out that there was already a function written which
handles collecting related articles together, and so the new code
is in fact shorter and cleaner than the old code.  

I have tested this in a variety of circumstances, but it would be good
for other people to test it and to examine the code, as my lisp is a
little rusty.  

The patch is attached below.  It replaces two functions in gnus-score.el.

Dan

-- 
Dan Christensen
jdc@math.jhu.edu