iahawks550 posted the IHME link yesterday. It's nice to see that at least the variance bars have tightened with the most recent update. I still think this study is **** ****, though, and I say that as a professional in the medical industry and a data scientist. They post their data, but they don't describe the statistical methods they use to arrive at their conclusions, there is no evidence of peer review (the "pics, or it didn't happen" test in my world), and as best as I can tell the information used by those models does not include intervention events. With a proper description, I should be able to recreate their results, but there is not enough model description for me to do so in practice. In my state (MA) their models is not tracking the reported stats, nor being adapted to current conditions (again, as best as I can tell, because the model is opaque).
For good hard data, I've found this site to be valuable: https://covidtracking.com/. The numbers I'm watching are confirmed case counts, hospitalizations, and deaths. Hospitalization/dealts are running ~10%/~1% of confirmed cases. If capacity (beds, ventilators, etc) is reached, then the death rate will go up. If capacity can expand, the rate will likely stay static. If the therapy trials prove to be effective, it will drop. The big question, that no one knows the answer to yet, is what is the rate of asymptomatic cases which require neither testing or hospitialization?
In short, I suggest one treat these models with skepticism. I think they are useful for predicting peak activity, but not total numbers. Don't despair or be led into fear by academic click-bait.
p.s. ICU cases in MA are running 20-25% of total hospitalizations [cite: from my employer's daily update]