Mean absolute error (MAE)
The average number of finishing positions by which the prediction missed, across the whole field.
For each driver, take the difference between predicted and actual finishing position, ignore the sign, and average over every driver in the race. An MAE of 3.0 means the model was, on average, three places off per driver.
RaceModel reports MAE with the full field and retirements counted. Dropping DNFs or scoring only the top ten makes the number smaller by design, which is why MAE figures from different sites are not comparable unless the scoring basis matches.