[Spambayes] Whitelisting for spam reports

[Richie Hindle]

[Manuel]
> I prefer to always train everything to keep my filter up2date.

[Tony]
> Research has shown, in every case, that this will give you inferior  
> results to training on errors & unsures only.

Whether you train on everything, or only on mistakes and unsures, doesn't
make a lot of difference to Manuel's argument.  Either way, he needs to
scan down the list and mentally filter out the spamcop replies - with
mistakes and unsures he wouldn't need to click the mouse for many of them,
but the fact that they're there makes life harder.

I don't think Manuel's request actually counts as whitelisting at all, and
I think it's quite reasonable.  What he wants (Manuel, correct me if I'm
wrong) is to filter out certain senders from the training list.  All
messages from spamcop would be automatically set to Discard and omitted
from the training list.  His email client's whitelisting rules would
ensure that they ended up in the right place.  Any messages that claimed
to be from spamcop but weren't would be problem - they would get through
the filters, and they would be untrainable, but as I understand it Manuel
doesn't have a problem with that (as the FAQ points out, the SpamBayes
developers see much more forged senders than ordinary users do).

When you look at it like this, I think it's quite a reasonable feature
request.  I'd appreciate this feature myself (although not enough to spend
the time to implement it, I'm afraid).  I get a lot of bounce messages
from the victims of spammers who are using my address as a forged From
address.  I wouldn't be comfortable training on them, because they are
pretty much indistinguishable from real bounce messages - I have to check
myself that the bounced address isn't one that I've sent mail to.  I could
use this feature to declutter my training lists.

> OK, my fault, i think i overrated the abilities of the developers.

...but Tony is dead right - this sort of thing doesn't help.

-- 
Richie Hindle
richie at entrian.com