That sort of forking model really doesn't work for prose. Making a fork is easy, but merging those changes with other forks to combine improvements rapidly becomes impossible -- especially if anyone tries to perform any sort of large change to an article.
There doesn't need to be a merge. Just pick the most promising version and edit it to create a new version. The old version sticks around. An add-only data structure.
How do you determine "most promising", though? Especially when many edits may not individually look like much (e.g, copyediting, fixing typos, etc), but help improve the overall quality of an aticle.
What this is likely to turn into, in practice, is a forest of dead forks. Division of effort, without a combination of progress.
