On Tue, 3 Jan 2017 10:39:52 +0100 Lukas Erlacher wrote: > On 12/28/2016 03:12 PM, RW wrote: > > > > It's done in spamd. Don't attempt to remove X-Spam-* headers > > yourself or it wont attempt to remove the mime encapsulation. > > > > I'd like to convince myself of that... I ran `sudo -u debian-spamd > spamc -c < spamspam.eml` on a mail that has spamlevel 14.4 and is > encapsulated in a spam report. It gave 2.7/7.0... which I suppose is > ok because it's an assessment of the spamminess of the whole mail. > But that doesn't convince me... > > How do I convince myself that it'll actually use the text of the > original spam mail to update the bayesian db? You could try what I just did.
Edit a spam report and put made up words at the beginning of each section, train it as spam and then put the made-up words through spamassassin -D bayes. printf "\n\n Lhjkl Ohjkl Ihjkl \n" | spamassassin -D bayes ... dbg: bayes: token 'ihjkl' => 0.986543689320388 Ihjkl was in the correct mime section at training - in the body of the embedded spam. What I don't get though is why isn't there a case-sensitive token "Ihjkl"?
