I've been looking for sudden jumps in the amplitude of the bendulum swings or in the clock's speed, but haven't seen any in many weeks. I've adjusted the run rate about as well as I can, and for the last 17 days, the clock consistently loses about 0.58 seconds per day or 6.7 parts per million.
I'm simply amazed that it works as well as it does. Unless something changes, I think it's time to tinker with a different design. I'm betting I can make a pendulum clock that's even better since a pendulum is more closely isochonous as the amplitude varies than is a bendulum.