Re: CrossAssassin results

[Blars Blarson <blarson@blars.org>]

In article <[🔎] 20040412131855.GA491@keid.carnet.hr> you write:
>What are the statistics if you use CA properly? :) IOW if you give it the
>thousands of messages that now go through SA?
>
>For purposes of testing, just log them, you don't have to actually kill off
>those 70-odd valid mails...

Looking through the crossassassinated files I found 17 more that
shouldn't have been killed, so I reinjected them (passed spamscan, so
they won't show up in the spamscan logs.)

$ egrep '^  ok .* ([4-9]|[1-9][0-9]+)$' /org/bugs.debian.org/log/spamscan.log | wc -l
     87

87 bugs that crossassassin would have caught improperly.

$ egrep '^  spam .* ([4-9]|[1-9][0-9]+)$' /org/bugs.debian.org/log/spamscan.log | wc -l
  66704

66704 spams caught by crossassassin (18 wrongly)

$ egrep '^  spam [0-3]\.[0-9]/4\.0 ([4-9]|[1-9][0-9]+)$' /org/bugs.debian.org/log/spamscan.log | wc -l
    251

251 caught by crossassassin but not spamassassin (18 wrongly)


    
All of these are since the spamscan.log format changed slightly when 
crossassassin started being used at about 2am Sat -0700.

Only the count at the time the message is seen is being used, I'm not
delaying all messages and checking again.  Doing so would increase both
the number of correct and incorrect kills.



-- 
Blars Blarson			blarson@blars.org
				http://www.blars.org/blars.html
With Microsoft, failure is not an option.  It is a standard feature.