Perhaps there could be some kind of semantic syntax that would group together the factual claim and supporting information (data and references). For example,
<"e=mc^2",cite1,cite2,data1,data2>
The claims still could be included in prose, but also easily extracted, analyzed, manipulated, etc.
With a little work the claims could form a semantic web, with all that entails, and the validity of cites could be checked automatically when the document is opened or on a regular basis, updating the claim appropriately. For example, the UI could display the claim differently based on whether the cited source has been withdrawn, partially invalidated, superseded, confirmed, rarely cited, heavily cited, based on impact factor, if the source's own citations have been invalidated (e.g., claimC cites claimB which cites claimA; if claimA is invalidated, there should be some sort of indication in claimC -- consider that the invalidation of a seminal claimA would set off a massive chain reaction, as it should), etc.
But that all seems obvious and I suspect I'm not the first to think of it.
I would prefer an approach like PaperBricks caught on instead: http://arxiv.org/abs/1102.3523 -- where a single section of a current paper could be published by itself.
I think it would be both amazingly useful and could speed up the iteration rate of science substantially.
What I find more of an issue is when trying to track down the source of some “fact” is going three or four papers up the citation chain and finding the paper at the end doesn’t actually say what the starting paper claims.
Get shit done and put it on your blog.
Normal people hear ideas, forget them, internalize them and regurgitate them as their own all the time. While you're scouring databases and calling libraries to try to find who's got access to the paper you're looking for because it looks like a primary source of what you're working on, you could be experimenting from secondary sources or even from textbooks if the thing is now mainstream enough.
If you want to be in Nature today, you need attractive claims and flashy graphics already. And the peer review is just not working, they are favoring already famous people and ideas.
Citation?