CORRECTION: I did a tweet thread yesterday which honestly went unexpectedly nuts. @arendbayer has pointed out that it contained a pretty serious mathematical error which I'd like to acknowledge and correct. Thanks to him for bringing it to my attention, apologies to you all.

The major themes of the thread remain correct I think: we are seeing linear decay of cases on a log scale, vaccination at a constant daily rate will cause quadratic decay, log scale graphs are the right tool. However, I overstated the effect this might have in my explanations.

It's possible that something like the blue curve may still hold, but this would be by accident because of more vaccinations being performed and each having more effect than I assumed in the thread, and I wouldn't want to take credit for being accidentally right.

The mistake is here: https://twitter.com/BristOliver/status/1353274729126387712
Essentially, the virus only reproduces every 5 days or so, so rather than e.g. a factor of (1-15c)(1-16c)(1-17c)(1-18c)(1-19c), you'd get a single factor of (1-19c). I've overcounted.

As a result, we'd end up with a (c/5) coefficient of the quadratic, rather than c, and the graphs end up considerably less optimistic than before based on the same (possibly conservative) assumptions.

Of course, I'm embarrassed to have messed up, and apologise again. It is no consolation that lots of smart people read the thread and didn't notice the mistake. However, science progresses best when people promptly acknowledge and own their mistakes, so I am doing that here.