The last notebook does include the code to test their model against the test dataset. However, students are encouraged to test their model against the test dataset themselves - to see how they do, then refit their models, then test again (iteratively). So, there's really not an opportunity for dishonesty on that end.
The only avenue for dishonesty would be submitting a fraudulent R-squared value in the google form. However, we have them submit their models as well. If we run it ourselves and find that their model doesn't produce the same R-squared value, then we flag the submission for dishonesty.
I hope that helps!