> > Why not, but I don't understand why you say this is less accurate.
> > DSPAM will continu to compute tokens as it was done before, and at
> > the end it will check if email address is in the whitelist and tag
> > the mail with "X-DSPAM-Result: Whitelisted".
> >
> That's not the way DSPAM works. Normally it checks at the beginning
> of the processing if a message is whitelisted or not. Not at the end.
> It would be less accurate if we would replace the current algorithm
> (using the whole From line) with a less accurate one (just using the
> "From" + "*" + "the email in lower case"). Maybe it would be wise to
> leave the current algorithm the way it is and introduce a new one
> (the fuzzy one) with own counters for when to tag a message as
> whitelisted?

Yes, this is what I meant.
(Sorry for my bad english, I'm french ;)

> 
> 
> > It is the same way of present whitelist, but management could be
> > done by the end-user.
> > 
> Well... using the same way as the current one would work but the
> managing part would be hart. This is because DSPAM computes a token
> out of the entry it want's to whitelist. This token is easy
> calculated but the computing from a token to a whitelist entry (in
> clear text) is the hard part. Adding would work very easy but
> deleting/changing would be the hard part since finding what token
> represents what whitelist entry would be +/- a unpractical task.
> 
> So from my viewpoint it can/(should) not be using the same algorithm
> as the current whitelist method. Probably more suited would be to add
> the user driven whitelisting in the same technique as the blocklist
> (flat ASCII text file). However... flat files are not known to be
> fast.
> 
> 
> > I talke about a manual intervention by the user.
> > I have 3 DSPAM servers running with Qmail, whitelisting on the MTA
> > level is not impossible, but with a centralized DSPAM database,
> > this is much more simple and clean.
> > 
> This is what I talk about. Using a central storage (database, file on
> a share, etc) would be a possibility. The reason why I think it is
> more wise to implement that outside of DSPAM is an economical
> viewpoint. It is more economical (less computing) to not start the
> content filter if you know at the beginning of the processing that
> you don't need to start the content filter. The only reason why I
> could think that implementing this in DSPAM could be beneficial is if
> you would train the content filter automatically with whitelisted
> messages. For example: 1) Message is from a whitelisted sender
> (well... can be faked but that's another story) 2) Message would be
> flagged as spam by the content filter if you would instruct the
> content filter to not look at the whitelist.
> 
> If this would be the case, then forcing the content filter to learn
> such a message as innocent would be beneficial to the whole filtering
> experience of the user. And then I would see a benefit in
> implementing the user driven whitelisting in the content filter. But
> if you just want to deliver a message and not pass it over the
> content filter, then it is more economical to implement that outside
> of the content filter. Why bothering in starting another instance of
> the filtering process when you anyway know that the message is
> whitelisted? Do you know what I mean?
> 

That's right, but in my case this is difficult to share data between my
different Qmail (simscan is using to run DSPAM).

> 
> // Steve

!DSPAM:1011,486a291a150926918273860!


Reply via email to