From context it's only used as a more concrete example of a product type to show how cardinality analysis works. I would infer that the idea is that this type would use mixedBit for the whole number and there would be a constraint that numerator < denominator. Fraction 5 1 2 would be the canonical way to represent 5 1/2 in that scheme. Practically speaking there's no reason to have the mixedBit field because a pair is enough to represent any fraction, but if it was simplified this way then the example would be redundant since there's already an example of a pair type on the same page.