The calculation is done based on the way spark code generates the partitions. BUt there can be around 5% difference in number as AQE sometimes takes in other factors such as partition coalesce