The rough math is based on % of negative vs. positive sentiment. There's likely a ton of noise around github that diluted the math. I can look into though, thanks for the feedback.