Improved Har Inference Using Power Kernels Without Truncation