spamassassin-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Warren Togami Jr." <wtog...@gmail.com>
Subject Re: Regarding Scoring of Mailspike
Date Mon, 12 Dec 2011 21:56:21 GMT
On Mon, Dec 12, 2011 at 11:26 AM, Warren Togami Jr. <wtogami@gmail.com> wrote:
> * Was setting these scores manually your response to my concerns about
> "reuse" and the difficulties we will face in GA rescoring?  I might
> even agree with this solution, although I believe it can be refined
> with further discussion.

Such high static scores don't leave much room for the GA rescoring to
balance the other DNSBL scores.  We would have have better
statistics-driven results if we let Mailspike's rules float, do GA
rescoring, manually adjust the results, then compare the before and
after fp-fn ratios to be sure it is sane.

1) First we need to consider the "reuse" issue I mentioned earlier -
decide if we will reuse or not depending on the status of the
participating corpora.
2) Fix the _BL composite score to 0.01 during the GA balancing,
allowing the _L's and _ZBI to balance along with the other rules.  The
results will NOT be linear, so the manual adjustment afterward is to
make it linear.  _BL remains as an informational rule in the final
scoreset that helps us to make an apples-to-apples comparison with
other DNSBL's.

The whitelist situation is more complicated than I have time to fully
write now.  I think we need to reconsider the entire whitelist
situation and set a consistent policy across all whitelists.  I
suspect you misunderstood my stance on whitelists ... I am actually
FOR whitelist, just we need to be careful about how they are scored.
The current DNSWL controversy regarding how they punish misuse is a
separate issue from how whitelists are selected and scored.

Mime
View raw message