Let's start by clarifying something. There are two uses for predictions: making decisions based on the knowledge that something will be X units by a specific da

Let's start by clarifying something. There are two uses for predictions: making decisions based on the knowledge that something will be X units by a specific date (e.g. buying stocks), or making decisions to avoid something becoming X by that date (e.g. planning interventions). Nearly all predictions of COVID-19 death tolls are the latter. They're meant to predict the outcome if all things remain the same so we can decide if something more needs done. If we respond to predictions by changing behavior, and the eventual number is lower, that's a good thing. The prediction worked. The model assumed no change, so you can't use reality after a change to dismiss the model. Your link is a discussion about a news report from March 11th, before the US and state governments started taking drastic actions. I'm not surprised the model predicted dire results and was followed by drastic action.