I don't have a solution to federated identity referencing (yet), but there are a bunch of historical and technical misconceptions here that I think might be muddying things a bit. I'm a bit pushed today so this is off the top of the head, rather than referenced - some timings might be a little off etc.
The standard character limit for SMSes in Latin alphabet is 160 characters. (I think it's 70 characters in Chinese - not sure about other languages.) Sometime in late 2006, Twitter changed their character limits to allow for additional flexibility. They had been reserving 5 characters for application commands, but they increased that to 20 to encapsulate usernames (which they simultaneously limited to a maximum of 15 characters). This is what gives us the 140 character limit. When building systems that similarly allow SMS communication, I think that's a pretty good model. The other major factor misconception I think folks in this thread have been missing is this: Twitter didn't invent the @ convention. They simply codified it (literally). Late 2006 - early 2007, my circle of folks on Twitter expanded to such a size that the nature of our communications changed. There began to be multiple discrete conversations at once (rather than one group conversation) and people started looking for methods to make sense of it. They adopted the @ reply convention (which has been used in many other places as an identifier - certainly Flickr and many forums I've seen. And email in the 70s, apparently - cool, I didn't know that!). In February or so of 2007, Twitter started linking those @ replies to the most recent tweet by that person. I can't remember if they started linking in-body references at the same time or not. The point is this: @ replies are not an arbitrary innovation by Twitter - they simply paved the cowpaths. I'd suggest we're pretty stuck with that format for now. Having said all of that, I'm currently most interested in the feasibility of individual aliasing รก la XMPP nicknames. If my client/service abstracts that alias for me, but uses the full XMPP address to communicate with other services (which then need to translate the address to the appropriate local alias), what are the problems with that? I already see a few, such as the one Bob mentions above with previosuly unknown IDs, but so far it seems the most workable option of those mentioned here. Lachlan Hardy
