This is by sleight of hand changed into a commitment to make it in two weeks, 100% sure. Then, 20% of time, developers are blamed for not fulfilling their committments or not estimating correctly. But they did!
If the most important thing for you is getting your commitments done with as high probability as possible, have a three-week sprint, but estimate only for two. Even this will fail once in a while, but less often. And maybe let developers work on the tasks of their choice in the remaining time. Autonomy, yeah!
"Two weeks on average" literally means it could be less but just as often it could also be more. If it is always two weeks or less, then it is not two weeks on average. What if it is always exactly two weeks, not a day more or less? No, that level of precision simply does not happen in real life.