David Goldsmith wrote:
> A messages that just made it through to my mailbox had the following SA
> headers:
>
> X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13)
> X-Spam-Level: ****
> X-Spam-Status: No, score=4.5 required=7.0 tests=BAYES_50,HTML_40_50,
> HTML_MESSAGE,URIBL_SBL autolearn=no version=3.1.0
>
> I bounced it to our 'spam' address and ran 'spamc' against the message
> and came back with:
>
> X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13)
> X-Spam-Level:
> X-Spam-Status: No, score=0.5 required=7.0 tests=BAYES_40,HTML_40_50,
> HTML_MESSAGE,URIBL_SBL autolearn=no version=3.1.0
>
>
> I've seen this often where email bounced by one of our users to out spam
> box appears to have a lower score when tested manually but in this case,
> I ran spamc within minutes of receiving the message.
>
> Any ideas on what may have changed in the Bayesian database in the short
> interval that would lower the confidence that the message is spam?
Define "bounced it to our 'spam' address".. What exact mechanism did
you use here?
I ask because Auto-processing learners are a dangerous minefield, SA's
bayes system is very sensitive to changes in:
From and To: headers
Body encoding
Both of which will be changed dramatically if you use "forward" on a
message.
