I definitely agree that custom storage engines can be scary! In a way you could say the fact we’re used happily by so many applications on literally hundreds of millions of devices is a test of our reliability, although it’s far from being a formal test of course. We field the overwhelming majority of our issues in public on GitHub and StackOverflow, so you can see the kind of issues we have there. We have a bunch of internal testing on top of that of course, including our own device testing, test suites, and fuzz testing. We’re always up for recommendations on testbench for the kind of storage we do (Jepsen concerns itself only with distributed failure scenarios iirc, which doesn’t apply to us since we’re an embedded/centralized database engine by necessity). I don’t know if this helps or not, but I’m always happy to chat if there’s anything more you’d like to know about