While Jepsen (and this article) is focused on behavior under node failure and network partitions, this caught my eye:
> It also exhibits Stale Read, Lost Update, and other forms of G-single in healthy clusters
This looks like quite a fundamental issue.