It looks like we're updating the stamp for t/countall while another task is replacing the file, which suggests a race condition in our state management database.