Re: [marf] New Version Notification for draft-li-marf-not-spam-feedback-00.txt

["J.D. Falk" <[e-mail]>]

On May 13, 2011, at 12:57 PM, Barry Leiba wrote:

> http://tools.ietf.org/html/draft-li-marf-not-spam-feedback

I agree that this fits nicely into the MARF WG.


The main comment I have on the document is that I think it needs a minor shift in attitude.  Statements like "It can be used to report a message that was mistakenly marked as spam" sound very certain, and that carries with it an expectation that the report will be acted upon.

I'm aware of four anti-spam systems which use "not spam" feedback today: Yahoo! Mail, Vipul's Razor, Cloudmark, and GMail.  (I worked on the first from 2004-2007.)  All of them take the "not spam" reports, and mix them with spam reports and other data using a proprietary algorithm.  In no case is a "not spam" report an immediate, automatic override.

(This differs from the "not spam" button in some MUAs, where it /can/ be an override -- but only for that individual user.  For example, the Yahoo! Mail MUA does this.)

The document does acknowledge that anti-spam systems can and should make their own decision (in sections 1.1 and 4), but I think that needs to be spelled out more clearly to avoid creating inappropriate expectations.


This proposal also brings up the more general question of feedback from an MUA to the user's mailbox provider, which I think would be a wonderful topic for an Applicability Statement.  But first, someone needs to actually implement it and try it out.

--
J.D. Falk
the leading purveyor of industry counter-rhetoric solutions