My definition of an email reputation collector is a service that gathers and processes data about email messages and/or senders.

I see reputation collectors as predictive or reactive:

 - Collectors that track "who is bad" by collecting data about who has already sent spam are reacting after substantial damage is done.
 - My favored predictive collectors focus on factors that can be measured before the message is accepted, even before the first spam from a new run is accepted. I also like predictive collectors that focus on positive or "good" characteristics rather than characteristics of "bad" senders.

I suggest using reactive collection data in a supporting role to predictive data. I believe putting reactive data on the front lines alone will result in higher false negatives.

For example, in the Outbound Index email reputation aggregator,
data reported by predictive reputation collectors such as domain age, relationship of the envelope-from domain to the sending server, and history of stability are known prior to accepting a message.

Senders who use their own domain names and their own servers and stay in one place would normally get a passing score. But data reported by reactive reputation collectors is used to tag senders in this category  who are stable, staying in one place and continuing to send spam using their own servers and their own domain names.

Some examples of what I would call reactive reputation collectors: 

 - spamtraps

 - human ethical judgements made after analysis of complete messages

 - human analysis of opt-in / opt-out policies

  - centralized unsubscribe request handling and analysis

   - legal system actions and convictions for internet abuse.