On 10-01-17 07:07, Michael B Allen wrote:
> If I understand correctly, the BAYES_X tags add a value corresponding
> to the X value. So BAYES_99 is basically adding 0.99 to the spam
> score?
This is incorrect. The number in the tag only corresponds with the
result of the bayesian classification. The score of the rule is static
and is set in one of the config files managed by sa-update.
>
> Ideally I feel it should be possible to scale this value such as by
> using simple multiplication or even exponentially.
>
> Is it possible to increase the score associated with the BAYES_99 and
> BAYES_999 tags?
Yes. Just define a custom score in local.cf for the rule:
score BAYES_99 3.0
Just be aware that the defaults are chosen wisely, and if you think some
score should be higher, then discuss that problem here. Maybe there's a
flaw in your setup that makes other rules perform less then optimal (DNS
issues f.i.).
>
> PS: Is it possible to see what values are associated with all tags for
> debugging purposes? Meaning can I run a command that dumps a list of
> all tags and their associated values so that I can decide which tags
> could have their scores adjusted?
You can grep for the scores in /var/lib/spamassassin. Note that the name
of a rule doesn't say everything: before tinkering with the score you
should verify that the rule is doing what you expect from it by reading
the regex.
>
> PS2: Is there a tag that indicates that the message contains a large
> amount of non-latin1 text? I do get a lot of legitimate non-ISO-8859-1
> messages but usually it's just a name or at most an address. So less
> than 100 bytes.
>
Please start a new thread and show us a sample of such a message, and
the scores you are seeing with your setup.
Kind regards,
Tom
