bayes training with whitelisted addresses

[Will Yardley <sa...@veggiechinese.net>]

How smart is SpamAssassin when dealing with whitelisted / blacklisted
email addresses and the bayes module? For instance, is it necessary
to retrain email which is spam (and would have been marked as spam
other than the whitelisting) if the sender or recipient address is
whitelisted?

Re: bayes training with whitelisted addresses

[Matt Kettler <mk...@evi-inc.com>]

At 05:49 PM 1/18/2005, Will Yardley wrote:
>How smart is SpamAssassin when dealing with whitelisted / blacklisted
>email addresses and the bayes module? For instance, is it necessary
>to retrain email which is spam (and would have been marked as spam
>other than the whitelisting) if the sender or recipient address is
>whitelisted?

The bayes autolearner intentionally does not consider the white/blacklist 
settings, because doing so would mean an accidental error in your whitelist 
settings could cause your bayes DB to become heavily poisoned by spammers.

ie: the common mistake of adding "whitelist_from *@veggiechinese.net" would 
very quickly result in a lot of spam being learned as ham, as many spammers 
forge your own email address, or another in your domain, as the sender.  In 
order to fix the problem you'd probably have to wipe your whole bayes DB 
and start over again.