Props to Tao for his response: he didn't just patch the sign error, he went back to Hildebrand's paper and found a better fix through log-concavity of the Dickman function.