[newacct583]

There are some bayesian tripwires in that analysis, though I think the data is good and it fits roughly with my intuition that bulk of covid deaths are in the "basically healthy and will live for many years" population.

The fact that a large-ish fraction of the entrants to the care homes die soon means that the people who don't die soon are going to be over-represented in the population. So if you pick a bunch of people from these homes at random, you probably (depending on the total population and total number of such deaths, which isn't cited) don't find people at the edge of death.

So I buy the upper bound, but think the actual number is likely to be much lower.

Stated more intuitively: there just aren't that many people "within months of death" at any time, and there are a lot of people dying to covid.