I think off and on about publishing a journal. I would do it this way: submissions would be accepted only if they had already appeared online and received widespread recognition. The articles could be submitted, nominated, or simply found by the editors. On acceptance, the article would be cleaned up, if necessary, and added to what amounts to a library of canonical publications. Regular 'bound' versions would be distributed to people outside the field. The idea of the journal is not to act as a content filter, but rather to confer, post-publication, recognition to works of particular merit that pass scientific scrutiny by a community of readers. My system is a bit simpler than Leigh Blackalls, and one day I will put it into practice - I simply need a setup like Aaarg.org (but without the login). It would double as a textbook as well.
Today: 0 Total: 27