If a message is submitted as spam, we consider the source and the content. We ignore questionable submissions, those that look like errors, and those that are likely to cause others false positives. (If you have a chronic local problem and want a specific black rule then contact support@).

Messages sent to spam@ are assumed to be spam or potentially hostile. We filter all of these messages through our rulebase, then each message is pulled up on our "Spam Hud" for review. The SPHUD gives us extra information about the email and helps us avoid errors and capture important patterns more easily.

Each message that survives is manually reviewed and considered for filtering based on our current policies and procedures. Each source that we can identify may be treated differently based on the assumed risk or probability of error that we have while reviewing those messages. For example, we have some folks who seem to be subscribed to everything there is on the 'net - and they submit every message they get - so we must avoid many of their submissions. We have others who only send us definite spam and so we take a hard look at every message submitted.

Of the messages submitted to our spam@ address, there are few that we don't add to the core system. However, if there is any doubt about the "spamminess" of the message based on the current trends of our subscriber base then we will hold off adding the content to the core rulebase.

There are some cases where we will attempt to unsubscribe an address from submitted spam. These are generally cases where spam has been submitted from an end-user on one of our client systems through an alias and where the unsubscribe process appears to be legitimate.

Our research teams continuously go through all of the messages that reach us through our spam@ address and our spamtraps and evaluate the messages to see how they can be filtered, and in some cases *if* they should be filtered. In order for us to generate filtering rules for our standard database we must determine among other things that:

• The content is likely to be viewed as unwanted by the vast majority of our subscriber base. Ultimately our definition of "spam" is dependent on the preferences of our subscribers - not our own opinions.
• The message contains content that can be clearly filtered using our current pattern matching technology with a minimum risk of producing false positives.

It is common for one system to receive some spam that other reporting systems and our spamtraps won't see. The more spam we have submitted to us the more comprehensive our filtering system will be. We continue to improve as our user base grows. All incoming spam is filtered through the current core rulebase so that we can concentrate only on any new messages that aren't being tagged.