On Wed, 2014-09-17 at 14:54 +0200, Reindl Harald wrote: > Am 17.09.2014 um 14:43 schrieb RW: > > On Wed, 17 Sep 2014 11:50:43 +0200 > > Reindl Harald wrote: > > > >> what i want to achieve is 4 levels of negative score for > >> both - FROM and TO just because it makes sense to handle > >> some mailing lists different without whitelist them completly > >> and the same for different RCPT, one user is living in asia > >> and needs complete different settings > >> > >> there is also a need to handle "securityfocus" ML different > >> to "postfix-users" without completly whitelist them > > > > A lot of people don't put mailing lists through Spamassassin, most of > > them have already been spam filtered, and to get the best results you > > have to extend your internal network and maintain it. And in my > > experience most list related spam is off-list anyway. > > that's a mail gateway delivering to different destination servers > for a bundle of domains - the problem is not list-related spam, the > problem is get a high score for discussions about spam which you > don't want to feed as positive into bayes
Just a thought - maybe a config setting to not do automatic bayes training for a give from/to addr would be more appropriate? Say a meta rule (more a "flag" but I don't know enough SA rules to know if those exist) that you set in your own rules. Eg. disable all automatic training for these addrs: no_bayes_from whate...@list.com no_bayes_to ot...@list.com That would be reusable in other rules/places if appropriate. Jesse > don't get me wrong but i spent really a lot of hours to consider > the environment which is nearly perfect and designed around 10 > years expierience with the userbase and other solutions > > just some wishes for extensions i would like to be able to > do myself and all about scoring "whitelist_from" is not > enough or goes too far > > SA is only a very small piece of the complete system > > >> finally offer in the existing webinterface (self-written) a drop-down > >> for each listed address/domain as well as define the score for each > >> level in the general settings to adjust > >> > >> as usal the cronjob builds "local.cf" from a template and that > >> settings > > > > If you are going to autogenerate them into local.cf you can generate > > header rules just as easily. The chief advantage of what you are > > asking for would be for users to set in ~/.spamassassin. > > running as milter with a milter user and it's global bayes > there is no users setting - just a global system maintained centralized > > >> score USER_IN_MOST_SPAM_FROM -4 > >> header USER_IN_MOST_SPAM_FROM > >> eval:check_from('USER_IN_MOST_SPAM_FROM') > >> > >> score USER_IN_ALMOST_ALL_SPAM_FROM -6 > >> header USER_IN_ALMOST_ALL_SPAM_FROM > >> eval:check_from('USER_IN_ALMOST_ALL_SPAM_FROM') > > > > The terminology seems a bit defeatist. Would you really let through > > almost all spam at a threshold of 11.0? > > forget the terminology and the score points > > the point what i am asking for is removed in the quote: > > * implement several "WL" score levels based on sender > * implement several "WL" score levels based on RCPT > -- Jesse Norell Kentec Communications, Inc. 970-522-8107 - www.kci.net