I have no idea what you think we’re disagreeing about. I’ve watched Cliff’s presentation a handful of times over the past decade and implemented something similar (albeit simpler) myself. Yes, the particular FSM in his hash table is interesting and novel, but just as insightful to me was his argument that data structures implemented as state machines built on atomic CAS can save you much of the trouble of proving you’ve put memory barriers in the right places. “All” you have to do is demonstrate that every possible state is valid.