Formulating the prioritisation question as a 'knapsack problem' and running some simulations, to conclude that WSJF can be improved by 1-2% on average and sometimes up to 30%! (in terms business value per sprint).
In this small study we define the greedy Vs Branch&Bound functions, simulate the inputs to get as close as possible to what may happen in product management with multiple/heterogeneous features requested each sprint (with Fibonacci scoring), and run a performance comparison montecarlo-like.