"autolearn=" and spams getting through.

[Steve Barnes]

Hi

Does anyone have know why this is occuring in maillog:

May 29 06:14:53 mail MailScanner[397]: Message 3C1731164B.AB9BD from
189.69.109.196 (efhcrewz_1995 at hillview.org) to domain.com is not spam,
SpamAssassin (not cached, score=
0, required 6, autolearn=)

In particular, the "autolearn=)" truncation. I've had a couple of spam
e-mails get through that were scored at 0.00 (and oddly enough, at
28.7 when later passed by hand to spamassassn -D on the command line).
The "autolearn=)" truncation occurred in both cases where the spam got
through. The only reference I've found via Gmane search dated back to
2004 and didn't seem relevent (or answered).

I have plenty of other entries that look fine:

May 29 00:05:36 mail MailScanner[97129]: Message 0B56A11638.A209B from
196.200.16.19 (sales at altn.com) to domain.com is not spam, SpamAssassin
(not cached, score=0, required
 6, autolearn=disabled, KHOP_RCVD_UNTRUST 1.00, RCVD_IN_JMF_W -1.00)

Could this be caused by a bayes database rebuild at the time of
spam-checks? I'm not really sure where to start looking since the
by-hand SA run set the spam on fire in both cases...

Thank you

Steve

PS, I'm using:

MS 4.76.24,
Perl 5.10.0,
SA 3.2.5,
FreeBSD 7.2,
Postfix 2.5.6 and
MailWatch 1.0.4.