metadata in learning_to_simulate

Hello, I want to know how vel_mean, vel_std, acc_mean, and acc_std are calculated in metadata? I used one of the datasets and used the sequence of positions in one of the segments to calculate mean and std, but the calculations were much different from those in metadata。
For example, I calculated a piece of data under Goop-3D:
Velocity Mean: [-3.57328663e-04 3.27340340e-05 -5.20929130e-04]
Velocity Standard Deviation: [44.02116085 17.71678278 38.50584431]
Acceleration Mean: [ 0.37627511 -0.23238154 -0.57692656]
Acceleration Standard Deviation: [28586.76965719 11446.06644121 24939.41377862]
I don't know if I made a mistake in my calculation