A model is as good as the data you have available, which is where abysmal testing in the US comes into play. Also, iirc, these projections are for the best-case scenario going forward, assuming strict observance of the social distancing, stay-at-home rules, etc ...
Not to mention that repeating...