I've read the white paper and used the equation EXP(u)/(EXP(u)+a-1) to get a probability score. I then rescaled the values to sum to 100. I assume that is correct.

That said, how do I rescale a combined P(ijk) where I use U(ijk) in the conversion from zero-centered to zero-based?

In my most recent study, the combined share of preference for my top 3 items was 44% but the unrescaled (but converted to zero-based) is .60...