William R Thomas wrote:
Yeah, but it's still going to score 12.0, so it's still going to be spam, and it's well over double the default tag threshold. You're not "rewarding" the spammer unless they manage to avoid being tagged.I had some questions about how the auto whitelist score was generated. Specifically, it seems like it is working opposite to the way it should be in some cases.
The equation for the AWL score is: ( meanscore - currentscore) * factor ^^^^^^^^ generated from historic messages ^^^^^^^^^^^^ the score before AWL ^^^^^^ Weighting factor default 0.5
I am seeing a decent number of messages where the AWL causes spamassassin to become more neutral. For example:
Spammy message (Score 14) from a historically spammy sender (Mean
10). This would get a AWL of -2.0. So it is essentially rewarding
spammers for sending a message that is more spammy than they have been
sending.
Also keep in mind you've got the opposite effect with the AWL too. You'd wind up punishing that same spammer if he sends one with a score of 6.0.
AND, keep in mind that although they get a credit for this message, that's going to end up bumping up their overall average, so future messages will be more likely to get punished. The only way a spammer can keep getting negative adjustments is if his spam scores keep going up, and that doesn't help him any. In fact, that's really what you WANT to happen. You want SA to get better and better at giving him high scores.
If the AWL is dishing out positive scores to spam, in some ways it's a bad sign because it means SA is less effective against a particular spammer than it was in the past.
We can also look at a good message (Score -5) from a historically good
sender (Mean -2.5). This would get an AWL of 1.25.
Which still ends up as -3.75.
I really think you're over-analyzing small changes in the value of a score. A change of 1.25 in a message that would other wise be 10 points below the spam tag level is pretty small. Now it's only 8.75 points below the spam tag threshold, and it's still way below zero. It's hard to call that a "punishment" when the end result is something so clearly way below the spam threshold.
And again, their average will go down, so future emails are more likely to get a negative adjust instead of a positive. And again, the only way to keep getting positive adjusts is for the scores to keep getting lower, which isn't a bad thing.
