This paper looked at how accurate the IHME model's predictions were and found that, even just one day ahead, the posterior interval of the model didn't contain the actual number of deaths in the vast majority of states. arxiv.org/abs/2004.04734 
Who could have thought that an ERF error function whose parameters are estimated by obviously misspecified models trained on poor quality data would not result in great predictive power? medrxiv.org/content/medrxi…
Not many "experts" apparently, but you're not supposed to say that, because it's populism and populism is bad. Allowing decision-makers to be guided by useless models, on the other hand, is not bad 🤷♂️
