Yeah, I actually thought of that first, and then discounted it for the reasons you mentioned. The big difference is that for the random-line solution, we have no cheap way of determining the entire set size, where with a hash, scalar-keys is cheap and tells us what we need to know.