All models are wrong, but some models are useful.

I mentioned that the binary outcomes were a shortcoming. However, it's a pretty simple model and I'd say it's pretty reasonable for back-of-the-envelope estimates.

The bias I was more concerned about was that of the human(s) who set the parameters. That's why I used their least optimistic settings.

One parameter that I didn't play around with was the discount rate. That would have a huge effect, but they did choose a value that's very common for these kind of economic models.