Times were collected by running one-shot qualifying for each track. The average of the two fastest drivers were averaged together to give a representative time of how good the AI can be. In certain cases the averaging was easier than others:
In other cases the difference between times was more pronounced:
All data was stored in a spreadsheet before being downloaded as a .CSV file that a simple python script turned into arrays.
Out of the first 100 races, Marie Laursen qualfied on Pole more than 50% of the time, leading me to believe she has the highest overall of the fake drivers. The highest margin to second place was also achieved by Marie Laursen, with a time of over 800 milliseconds to second place.
On average, the difference between each difficulty level is around 0.013%, which equates to 0.0078 seconds difference per minute of laptime per difficulty level. This is easier to see on the longer tracks such as Belgium.