When we uphold a particular rule in an FP report it only means that according to the statistics we have at that moment- that rule is doing significantly more good than harm -- so, it follows that mitigating the rule in question must be a localized (not global) process.

In these cases, we offer to either block the rule or to compensate with a white rule. Often through the course of investigating such a case we end up removing the rule from the core rulebase and developing more specific alternatives.

In these cases, we never intend to imply that the rule is "legitimate" -- to the contrary: For you (the one reporting the false positive) it is clearly not legitimate. The question then is what to do about that.

The SNF rulebase system was designed from the outset to recognize that each system's definition of spam - and even the way in which they use SNF to filter their messages will be unique. Along the way we have been able to prove the theory that the vast majority of folks will agree on most rules and that the requirement to make localized customizations can be provided in a practical, sustainable way.

We have also largely proven that a combination of automated and human learning processes are capable of not only responding quickly to the filtering challenges presented, but they are also able to achieve a continuous improvement in accuracy and even the ability to improve over time.

Discovering and adapting to false positive cases is a continuous part of the process by design. There is no other way to hit a moving target - let alone a target that is more than one place at a time.