The core of the paper is a mathematical model of how to determine whether there is more value in improving predictions or more value in increasing the follow-up testing budget.

I think the framework they define makes sense and is mostly well justified. This is mostly quite a good paper.

There is just one place where I think the claims are at the very least misleading; maybe even wrong: the fact that and are treated as having the same "unit". There is a paragraph on page 5, column 2, lines 251-264, which admits that this is wrong to do; it admits that one should consider the costs of and instead, but then the rest of the paper treats and as comparable anyway – for example, in Figure 3, you draw a line at PAR=1 as if that were a meaningful threshold in any way!

I would think PAR should be defined like this:

Where is the total cost. This definition is robust to change of variables. For example if you were to go from to .

I understand why the authors didn't want to do it like this – it means having to model the total cost as well – but right now it just seems wrong to me.

I would also have wished for more of a direct contrasting with Perdomo (2024). Why make different choices than that paper?