> If they know they will get the reward then they know the exact opportunity cost and can weigh that against performing the behavior.

So you need to to randomize the value of the treat. Otherwise the dog knows the upper bound of the opportunity and the same reasoning applies.