Unfortunately, I was wrong about the consequence of the zero variance limit. I think I misinterpreted the paper I was reading. However, I've done some other calculations after realizing that I'm not ever sorting my data.

The covariance calculation became simpler, so I have fixed that now.