On Fri, 23 Dec 2016 10:12:54 +0000 Sebastian Arcus wrote:
> On the command line it is hitting BAYES_99 and BAYES_999 - while > through Exim it doesn't. ... > What could possibly account for the large discrepancy in bayes > results? If you're running spamd as the spamd user with a global database and both scans are using spamd that should rule-out the normal explanation of two separate databases. I don't know much about Exim but some Amavisd users have assumed that they are using spamd when SpamAssassin is being used as a library. Assuming that they are using the same database, perhaps the spam was autolearned when it was originally scanned from Exim, or other training changed the result. Another possibility is that the difference is due to additional headers added after the Exim scan. For example if you are using a mixture of autotraining and manual training and the ratio of spam/ham is much higher in the manual training, these extra headers may look spammy and skew the result on rescanned mail.