* Steve Lamb <grey@???> [2003-11-05 15:53]:
> Dominik Ruf wrote:
> > It seems this is definitetly not the case if running e.g. Spamassassin
> > at SMTP time which will return the same score (besides upgrades,
> > etc.) for the same message causing a defer until the end of
> > retry times of the remote site.
>
> This is not true. If it is an extremely marginal case subsequent
> deliveries may be let through because of a shift in the Bayesian score.
> Granted it is highly unlikely but it is possible.
Agreed. This would be an unlikely case (of maybe some more?) which
I meant when saying "besides upgrades, etc.". :-))
But I think these cases are all relatively unlikely and they don't
justify a defer.
Additionally, it's not that trivial (and IMVHO not that wise) to
implement Bayes together with rejections at SMTP time because one
could only use a shared Bayes database (-> not as well as per User
Bayes DBs) which also needs some smart solution to feed it correctly.
Dominik
