To properly support optimisation proofs, we should implement them by using assumptions rather than posting new hard constraints.