[Pearg] IP address Privacy Interim Follow up - Reputation System

[Matthew Finkel <[e-mail]>]

Hi Everyone,

I'd like to follow up on the topic of reputation systems we briefly
touched on Tuesday. I'd like to understand the criteria of a system that
can help the Internet move towards deprecating IP addresses as a
fundamental signal in account-security/anti-fraud/anti-abuse systems.

>From the perspective of a web user agent, I'll start with the following
properties:

  - A reputation and its associated identity must not be linkable
    - This excludes the user and, possibly, a "reputation provider"
  - A reputation must not be universally unique
    - I suggest bucketing reputation into ~5 sets
  - A reputation must not be a tracking vector
    - Two "proofs of reputation" must not be linkable, neither across
      sites nor on the same site.
  - A client must be able to know (or query) its recent reputation
  - A service provider must be able to increase and decrease a
    reputation
  - A client must only use a recent reputation
    - A reputation should become stale and invalid within a short period
      of time
  - A client must be able to audit its reputation
  - A client must be able to dispute changes in its reputation
  - transitioning from a "bad" reputation to a "good" reputation should
    be reasonably possible
  - Computing a reputation must be well defined
  - The system must support "anonymous" identities
    - A client must be able to create a new and unique reputation not
      linkable to previous reputations or identities
  - A reputation must not be influenced by an associated IP address

While these are not all hard requirements, it is important that this
system doesn't become a "virtual credit bureau" where someone's
reputation is based on black magic and good fortune.

As a meta comment, I'm relatively new to IETF/IRTF processes, so I
appreciate recommendations on how this discussion should proceed.

Thanks,
Matt