>>Yes, but if everything that SA considers not to be spam bypasses
>>TMDA why do we have TMDA at all?
>>
>>Sounds like SA will let some spam through if it judges wrong.
>>So why not let TMDA handle those?
>
> There are two ways SA can be wrong:
>
> 1. false positive -- ham seen as spam
> 2. false negative -- spam seen as ham
>
> You're talking about 2.  I avoid that by setting required_hits very low
> (-1.5), but you could just as easily remove the rule that allows those
> to deliver.  I found that messages SA allowed to bypass TMDA typically
> were spam, so I made its requirements much more strict.

I guess I am more conservative, we're talking about a very small
pecentage of messages, so it is not a burden on TMDA, since you have
observed they are generally spam anyway

> TMDA takes care of 1.  By giving spam a chance through (via whitelist
> or confirmation), you avoid the false positive problem.
>
> The other benefit we get from SA is that we can take things that are
> obviously spam (i.e., spams with very high scores) and NOT challenge
> them, saving time and bandwidth.  If you don't trust SA's idea of ham
> at all, you can probably still find some score at which you feel
> confident SA really HAS identified a spam, and skip confirming those.
> That alone is an important benefit.

I am not questioning the motivation just that one minor detail

tom


_____________________________________________
tmda-users mailing list ([EMAIL PROTECTED])
http://tmda.net/lists/listinfo/tmda-users

Reply via email to