Re: Whitelist or BAYES?

On 10/1/2024 8:58 AM, Bill Cole wrote:
>
> On 2024-09-30 at 16:22:49 UTC-0400 (Mon, 30 Sep 2024 16:22:49 -0400)
> joe a <[email protected]>
> is rumored to have said:
>
>     On 9/27/2024 04:05:51, Matus UHLAR - fantomas wrote:
>
>         On 26.09.24 10:27, joe a wrote:
>
>             Maybe I should not ask this, but . . .
>
>             A relatively innocuous member informational email from a
>             local town Library (monthly) gets marked as spam as shown
>             below.
>             The BAYES_99 and BAYES_999 values are something I am
>             toying with for other reasons.  Seems odd these should hit
>             either one of those tests.
>
>             So, on the one hand I can add them to whitelist and be
>             done with it, or I can add
>             them to missed HAM for re-learning.
>
>             Which is the best approach?
>
>         so far, both. You may need to relearn multiple their (monthly)
>         mails before it has effect.
>
>             X-Spam-Report:
>                 *  4.1 BAYES_99 BODY: Bayes spam probability is 99 to
>             100%
>                 *      [score: 1.0000]
>                 *  5.0 BAYES_999 BODY: Bayes spam probability is 99.9
>             to 100%
>                 *      [score: 1.0000]
>
>         You have raised BAYES_99 and BAYES_999 to huge values so I
>         recommend to rethink that.
>
>     You some "don't because" examples?   Seems to me, off hand, that
>     if it's 99% or 99.9% then a high value does no harm.  Perhaps half
>     what I have would be sufficient though.
>
> Bayes is a statistical method and so will always make some errors, as 
> in this case. BY DEFINITION, one in a hundred messages hitting 
> BAYES_99 will be ham, as will one in a thousand that hits BAYES_999.
>
> I can't claim that the default scores are the best possible ones, but 
> they don't result in many false positive *final scores* for most people.
>

Also, keep in mind that BAYES_999 is an add-on to BAYES_99.  Any message 
that hits BAYES_999 will also hit BAYES_99.  That is why the default 
score for BAYES_999 is only 0.2.

The way you have your scores set will ensure that any message that hits 
BAYES_999 will get 9.1 points added (4.1 + 5.0).  This may or may not 
work for you, but you should be aware of it.

-- 
Bowie