There's one thing I'd add here, as advice to potential implementers of collaborative editing algorithms. Test the system rigorously, autogenerating concurrent w

There's one thing I'd add here, as advice to potential implementers of collaborative editing algorithms. Test the system rigorously, autogenerating concurrent workloads of all the operations (including undo), and also various delays. If this had been done early on, all the broken algorithms (some published with formal proofs of correctness!) would have been caught. A lot of the broken algorithms work under light loads - TP2 requires at least three concurrent edits, two inserts and one delete, to trigger.