I'm not entirely sure I understand what you're saying here, but I think the difference you're noticing is due to co-variance, and the fact that this calculator doesn't take it into account (or just assumes that it's 0).

For example, when you use covariance properly you could say: x=2+/-2 x-x=0

since covariance = 1

But this calculator assumes that covariance is zero, so it does the calculation (2+/-2) - (2+/-2) = 0+/-sqrt(2^2 + 2^2) = 0+/-2.8

If I'm misunderstanding and this is another problem, please help me, I'm not all that good with uncertainties, but am trying to learn more in this area as it's very useful.