This article makes the mistake of assuming the naive error bound is the actual error on the specific problem. This problem has very smooth functions, infinitely differentiable, and has derivatives tabulated. The error bound he states doesn't use this information.
The correct answer is to check this problem specifically, not apply much looser bounds.