You could use the definition V(start) = Q(start,a) where a is the action recommended by the final learned policy.

However, the final Q values may not be perfectly accurate, so it is better to do what the project description says: use policy evaluation to measure the goodness of the final learned policy.