[EMAIL PROTECTED] wrote: > Hi gurus, > > Recently, I've upgraded to spamassassin 3.2.0 called from amavisd-new. > I've seen that this version is more agressive, and for example it > detect as spam > a legitimate email with next score: > > X-Spam-Status: Yes, score=4.884 tagged_above=-999 required=3.5 > tests=[MIME_BASE64_TEXT=2.796, RCVD_BAD_ID=2.088] >
SA doesn't think this is spam (score < 5). _you_ decided it is spam. Rule score are computed using the threshold of 5. Using another threshold may work, but don't forget that the optimization problem that the perceptron solves is not linear. if a lot of spam is missed, don't lower the threshold. Instead, find out if you can add rules to detect that spam. > Is this correct? > > I know that, perhaps, email was'nt sent correctly, but what can i do? > > - increase required score until 6 or 7 for example? > - avoid these tests: MIME_BASE64_TEXT, RCVD_BAD_ID or > score with lower punctuation depends on how many legitimate mail you get that trigger these. > - other solutions, > use Bayes and train on errors.
