As always, many of these quantities are gross approximations, modeled as continuous improvement instead of step functions, and often missing out on where perfor

As always, many of these quantities are gross approximations, modeled as continuous improvement instead of step functions, and often missing out on where performance has permanently plateaued. Looking at the numbers for any one year is a waste of time. Only the long-term trends are instructive, particularly where operations trade place in rankings (eg. due to bandwidth increases with constant latency). There's also something fishy about the scrolling in the source code box.