The stable timestamp in test/format is set to the current timestamp, it really should lag the current timestamp, and ahead of the oldest timestamp.

Also there's no reason that the stable timestamp shouldn't move unconditionally (currently it only moves if return_to_stable is enabled).