Partial messages (usually split by SMTP limiters) can
have base64 data without appropriate MIME header,
therefore all data will be tokenized and will most
likely result in numerous
'skip:q 70': 0.98; 'skip:g 70': 0.98;
'skip:j 70': 0.98; 'skip:o 70': 0.98; 'skip:v
70': 0.98;
'skip:k 70': 0.98; 'skip:s 70': 0.98; 'skip:t
70': 0.98;
'skip:a 70': 0.99; 'skip:n 70': 0.99; 'skip:p
70': 0.99;
that will raise its spam score to the roof and cause
the message to go to 'unsure' if the sender is
well-trusted, or to 'spam' otherwise, where it can be lost.
The classifier would do a better job by caching the
classification of messages
containing a header like
Content-Type: message/partial; total=2;
id="01C50EFC.CA6C2680@quake"; number=1
and applying it to the subsequent parts.
Logged In: YES
user_id=790676
Or, simpler, treat the body of continuation messages as
"empty" .
Logged In: YES
user_id=552329
Sounds more like a feature request than a bug to me, so
changing.