I've noticed that getting stats on what spamd is doing is not so easy, and 
involves using 3rd party log-filters and report generation mechanisms.  Some 
tools are specialized for spamd running on a host separate from the mail 
server, others work when spamd is on the mail server, others don't care (?).  
Per a discussion on the spamassassin discusion list, I'm entering this as an 
issue for developers to consider, and for users to communicate 
preferences/ideas/requirements to help give the developers a feel for what 
people really want

Obviously, spamd can only report on the mails it sees (mails junked by 
procmail rules before spamc is invoked etc... are invisible to spamd 
obviously).  

Despite that, it would be immensely usefull if spamd could do a little book 
keeping and provide some basic metrics like: I saw X mails in the past Y 
hours, and Z (zz%) were spam.  

Some ideas for requirements/features:
* Ability to specify list of emails to receive report.
* Ability to select which metrics are of interest. 
* Easy to read, or easy to parse formats (or both?) 
* Ability to specify frequency of reports
* Ability to specify window for which stats are kept. 
* Persistance of metrics/counters across restarts with minimal dataloss for 
rude crashes or killings. (hourly dumps to disks?) 
* Ability to dump data into a relational database (somewhat advanced, but just 
give people the data and let them decide what they care about?  The user 
community could share canned sql queries perhaps?).  Maybe the best approach? 
* Ability to turn metric generation/tracking on or off.

The spam assassin community can probably come up with better 
ideas/requirements about what the metrics could look like.  It seems that once 
Spam Assassin does it's checks and makes its decision regarding the 
disposition of a mail, it would be pretty trivial just to increment a few 
counters, or dump some data.  Post-processing logs seems like a backwards way 
to get this information.