[hwayne]

With one client I have, we know the TLA+ model is accurate because we're extracting tests directly from the spec. It's kind of a riff on what MongoDB does in this paper: https://arxiv.org/abs/2006.00915