On Nov 29, 2009, at 12:40 AM, Alessandro Vesely wrote: > > SA provides a certainty measure, and a server may make different decisions > according to that score. Possible actions may include: > > * set message header (SA default, Authentication-Results), > * set IMAP keywords (either Junk or NonJunk for TB users), > * deliver normally, > * deliver to a Junk folder, > * reject with various levels of detail, > * drop or quarantine, and possibly even > * create a record for a class of of messages (based on envelope data) that > the recipient can whitelist using a web form. > > Is there an obvious decision matrix? In particular, would reliability be > better if a server rejects messages with higher spam scores and delivers to a > Junk folder for medium-to-hight, or the other way around? >
Depends on your goals, and on your user demographic. And your definitions. I remember a study from a few years back that said that the majority of typical (consumer or non-technical business) users will only look in a bulk folder for about two weeks after they're first exposed to it. With that sort of user base then you're more likely to lose wanted email if you deliver it to a bulk folder than if you reject it, as 9 times in 10 delivery to bulk folder is indistinguishable from silently discarding. On the other hand, more sophisticated users (a small minority but the sort of users who might handle role accounts, say) are used to the concept of bulk folders, and have tools like full text search available to them so delivering to a bulk folder may cause less loss of wanted email to those users. It'll likely cause delivery of more unwanted email to their bulk folders, but they have the MUA tools to deal with that tradeoff. (Dealing with presentation of spam filtering decisions is a subset of the more interesting general problem of assisting recipients in handling mail volume - prioritization, tracking and so on). Cheers, Steve
