[Gllug] More Viagra

[Nix]

On Sat, 22 Mar 2003, Adam Bower said:
> On Sat, Mar 22, 2003 at 03:34:15PM +0000, Nix wrote:
>> For me, it fired NO_REAL_NAME and SUBJ_VIAGRA, which wouldn't have been
>> enough to class it as spam, except that it also hit DCC_CHECK,
>> RAZOR2_CHECK, RAZOR2_CF_RANGE_91_100, and BAYES_60. 6.9 hits in total ->
>> spam.
> 
> How do you find Razor? I am currently not using it because I don't trust 
> other people to not wrongly classify things such as weekly whats on at the
> cinema type mailings etc

Ah, well, I don't get many of those, and those that I do get caught as
nonspam by other SA rules or because I know I'm on that mailing list and
don't bother to spamscan it (but the latter's for high-volume lists only).

>                          and I am too lazy to add it to my spam assassin
> checks (and configure spam assassin to run directly on incoming mail rather
> than through procmail).

That's the only sane way to use Razor, I'd say; it's not *inaccurate* as
such but its accuracy level is low enough that you have to use it as
part of a wider suite of tests. The `is it present in razor' test (the
one that ignores the confidence factor Razor places on its opinions)
gets 2.029 points if Bayes is off and only 0.787 if it is on.

The mass-check runs show that while both are very good at zapping
stubborn FPs, a well-trained Bayes is *substantially* better at doing it
than Razor is.

So I use both. :)

>                         Does using razor increase false positives?

If you use it as a sole check, it might --- although I only ever test
mail that's not from high-volume mailing lists I know I'm on (basically
I test stuff destined for my primary mailfolder), so I get very few FPs
in any case. FNs are more of a problem for me but Bayes seems to have
killed most of them too.

-- 
#ifdef USE_ISPTS_FLAG
		} else {	/* else pty, not pts */
#endif

-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug