L2 vs. L2 error in Differential Equations Benchmarks

How exactly is the L2 error calculated in Differential Equations WorkPrecisionSet? Not much is mentioned about it in the manual. I understand the l2 error calculates the error at every time step taken by the integrator Then how many points in the tspan is used for L2 error calculation?

It approximates it using 100 evenly spaced points:

Thanks for the clarification.
Since the L2 norm is used here as a measure of global error shouldn’t the number of equally spaced time points depend on the scale of the ‘tspan’?

There really isn’t a good general way to do that