Why do we subtract background from raw before unfolding?

[vetlewi]

In the Ensemble class there is the option:

1. Provide the background subtracted gamma-excitation matrix
2. Separate prompt and background gamma-excitation matrices.

If separate matrices are given then perturbed prompt and background matrices will be sampled separately before subtracting the background from the prompt. This background subtracted matrix is then unfolded.
In the code:

if self.bg is not None:
    prompt_w_bg = self.generate_perturbed(step, method, state="prompt+bg", rstate=rstate)
    bg = self.generate_perturbed(step, method, state="bg", rstate=rstate)
    raw = self.subtract_bg(step, prompt_w_bg, bg)
else:
    raw = self.generate_perturbed(step, method, state="raw", rstate=rstate)
unfolded = self.unfold(step, raw)

I'm wondering why the we do the background subtraction before the unfolding and not after?
In code that would be:

if self.bg is not None:
    raw = self.generate_perturbed(step, method, state="prompt+bg", rstate=rstate)
    bg = self.generate_perturbed(step, method, state="bg", rstate=rstate)
    raw_unfolded = self.unfold(step, raw)
    bg_unfolded = self.unfold(step, bg)
    unfolded = self.subtract_bg(step, raw, bg)
else:
    raw = self.generate_perturbed(step, method, state="raw", rstate=rstate)
    unfolded = self.unfold(step, raw)

I can't really see any reason why this isn't valid. I would think that this would be a convenient way around having to remove negative counts before unfolding as neither the raw nor the bg matrices will have negative counts. Maybe there is something I'm missing?