%ents; ]>
Entity Reputation This specification defines an XMPP protocol extension for communicating the reputation of any entity on the network. &LEGALNOTICE; 0275 Experimental Standards Track Standards Council XMPP Core reputation &stpeter; 0.1 2010-01-21 psa

Initial published version as accepted for publication by the XMPP Council; defined XML schema.

0.0.1 2010-01-11 psa

First draft.

Reputation systems are used in many online communities to increase trust and to encourage communication, commerce, and other forms of interaction. The public XMPP network might benefit from instituting a reputation system for servers, for end users, or both. The benefits might include fast blacklisting of rogue servers and other bad actors, differential quality of service based on reputation, delayed entry to &xep0045; rooms for low-reputation users, integration with &xep0016;, and the like.

The following terms identify the entities mentioned in this document:

The approach taken here is that an XMPP entity (a "Subject") is innocent until proven guilty, so it starts out with a score of zero. Good behaviors will increase a Subject's score (up to a maximum of +127), whereas bad behaviors will decrease a Subject's score (potentially down to a minimum of -128). Any entity (a "Relying Party") can query another entity (a "Vouching Party") about the reputation score of a Subject. A Vouching Party might be a fellow IM user (e.g., a buddy in one's roster as defined in &xmppim;), the server to which a client connects, a fellow server to which a server connects (see &xep0267;), or a specialized reputation service (similar to a DNSBL on the email network). Changes in reputation scores can also be sent in real time to Relying Parties, either as part of &xep0268; or as separate notifications.

To determine reputation in an objective way, it is important to define the specific behaviors that are used as measurable dimensions of good or bad reputation. The following sections attempt to do so for XMPP servers and XMPP users, with some rough point values.

On the theory that it is more important to reward positive behavior than to punish negative behavior, we first define a number of criteria for increasing the reputation score of an XMPP server (naturally this list of criteria is not meant to be exclusive).

Criterion Suggested Point Value
Presents a certificate issued by a recognized certification authority. +15
Requires CAPTCHAs or other hurdles for account registration (see &xep0158;). +5
Supports XEP-0268: Incident Reporting. +5
Supports reputation scores for its users (i.e., this protocol). +5
Requires use of Transport Layer Security (TLS) for client-to-server connections. +5
Provides the _xmpp-client DNS SRV record. +5
Provides the _xmpp-server DNS SRV record. +5
Provides a website with accurate information and contact addresses. +5
Service answers &xep0030; information requests sent to bare JIDs, including identification of admin accounts and anonymous users. +5
Administrator answers email sent to mailto:xmpp@domain.tld (see &xmppcore;) and takes appropriate actions to solve reported issues. +5
Time online (e.g., based on whois lookup or known deployment of an XMPP service). +3 for each year
Admin factor (average of admins' reputation scores, divided by ten and rounded up). Varies

For example, a server that (1) meets all of the foregoing criteria, (2) has been online for 7 years, and (3) whose admins have an average score of 37 would have a reputation number of 15+5+5+5+5+5+5+5+5+5+4+21 = 85.

By constrast, a server that does not have a CA-issued cert, does not require CAPTCHAs for account registration, does not support incident reporting, does not support reputation scores, does not require TLS, does have SRV records (+10), has no website, does not answer service discovery requests about its users, has not verified the xmpp@domain.tld email address, has been online for 1 week, and whose administrators are unknown would have a reputation score of 10.

The reputation associated with an XMPP account (typically but not always a user) is harder to quantify and easier to fake than server reputation. The following are some possible criteria.

Criterion Suggested Point Value
Account has service discovery identity of account/admin +15
Account has service discovery identity of account/registered +5
Age of account +5 for each year
Verified email address +5
Verified website +5
Password strength Measured on a scale from 0 (e.g., "password") to 5
Reputation of buddies known to server Divide average reputation by 10
User has PGP key, X.509 certificate, or other public key +10
User has passed a CAPTCHA test (e.g., during account provisioning) +5
Chatroom ownership / administration For each room owned, divide room's reputation by 10; for each room administered, divide room's reputation by 20 (e.g., +6 and +3 for a room with a reputation of 60).
Chatroom banning For each room in which the user is banned (XEP-0045 "outcast"), divide the room's reputation by 10 and decrement the user's score by the result (e.g., -6 for a room with a reputation of 60).
Rate limiting For each rate limiting incident, -5.
Incident reports For each validated incident report, -10.

For example, an account that is an admin of a server (+15), has been online for 5 years (+25), has a verified email address (+5) and website (+5), has a strong password (+5), has a "buddy reputation average" of 40 (+4), uses a public key (+10), has passed a CAPTCHA test (+5), owns 3 chatrooms with an average reputation of 30 (+9), and has not been banned from any chatrooms, rate limited, or been the subject of any incident reports would have a reputation number of 83.

By contrast, an account that is registered (+5), was just created, has no verified email address or website, has a strong password (+5), has a "buddy reputation average" of 10 (+1), does not use a public key, has not passed a CAPTCHA test, owns or administers no chatrooms, has been banned from 3 chatrooms with an average reputation of 30 (-9), has been rate limited twice (-10), and has been the subject of 2 incident reports (-20) would have a reputation number of -28.

Any entity can have a reputation score: servers, end-user accounts, chatrooms, chatroom servers, publish-subscribe servers, vouching services, etc. Criteria for entities other than servers and accounts are yet to be described.

In order to query a Vouching Party about the reputation of a Subject, a Relying Party sends an IQ stanza of type "get" containing a <score/> element qualified by the 'urn:xmpp:reputation:0' namespace &VNOTE;.

]]>

The Vouching Party would then return an error or a score.

]]>

Any XMPP error might be appropriate (e.g., &forbidden; if the Relying Party is not trusted at all by the Vouching Party).

To advertise its support for reputation scores, when replying to &xep0030; information requests an entity MUST return a feature for 'urn:xmpp:reputation:0'.

In order for an application to determine whether an entity supports this protocol, where possible it SHOULD use the dynamic, presence-based profile of service discovery defined in &xep0115;. However, if an application has not received entity capabilities information from an entity, it SHOULD use explicit service discovery instead.

Any entity can keep a reputation score for any other entity. Although some entities might be considered more knowledgeable than others, that judgment is up to the entity that makes a request for a reputation score and is not a feature of the network itself.

This document requires no interaction with &IANA;.

This specification defines the following XML namespace:

  • urn:xmpp:reputation:0

Upon advancement of this specification from a status of Experimental to a status of Draft, the ®ISTRAR; shall add the foregoing namespace to the registry located at &NAMESPACES;, as described in Section 4 of &xep0053;.

&NSVER;
]]>

Thanks to Dan Brickley, Kevin Smith, Mike Taylor, and Matthew Wild for their feedback.