Differences between revisions 3 and 4
Revision 3 as of 2005-09-01 12:25:39
Size: 2364
Editor: Andy Balaam
Comment: Minor typo: yor -> your
Revision 4 as of 2009-09-20 23:17:05
Size: 2364
Editor: localhost
Comment: converted to 1.6 markup
No differences found!

A non-spam message was flagged as spam! Can I send it to you so you can update the filters?

Yes, but you need to do some legwork to make this worth while.

Read FalsePositives. If any of the suggestions there apply to you, then follow those links and suggestions.

Check to see which rules hit the email, and their scores. If you adjusted any of the scores, or if you've added any custom rules or rule set files, or if you've lowered your required hits threshold from the default, then you need to reverse those changes. If the email would not have been flagged as spam without your changes, then there's nothing we can do to help -- the problem lies within your changes.

If you're running Bayes, make sure the email is learned correctly by the Bayes function, and then retest. If SpamAssassin correctly handles this email after training, then your system is working correctly. (The very occasional false positive is unavoidable.) If you get multiple false positives over a short period of time, with BAYES_90 and BAYES_99 scores, your Bayes database may have been corrupted by the accidental training of non-spam as spam. See BayesInSpamAssassin

If you have the AWL rule enabled, check to verify that its score is reasonable. Would the email have been flagged as spam if AWL had not been scored? If your AWL values are out of kilter, you need to fix those. See AwlWrongWay

If after all that, your non-spam still scores like spam, then it may be worth posting to the SA Users Mailing list (see MailingLists). Search the archives to see if your problem has already been reported. If not, then report the problem, stating clearly a) how you're invoking SA, what version of SA you're using, what major options you're using, and your platform. Assuming the email is reasonably small, then attach the email, with full original headers as you received it, to your report. (Please don't dump large emails onto the list without warning.) If there's an easy solution to your problem, we (the community) should be able to see it there. Never ever cut-and-paste emails, since that will almost always change the spacing and line-breaks of the message

If after that review it's determined you've actually found a scoring problem in SA, then you can formally submit the email to the development team (see UsingBugzilla).


CategoryBayes

DoYouWantMyHam (last edited 2009-09-20 23:17:05 by localhost)