Comment by stargrazer

which would be explained that those contribute very little to weighting, and so like extrapolating beyond the last end-point, errors accumulate significantly.