Following up on our discussion in April/May about the non-uniformity in COVID transmission event sizes -- i.e. the average reproduction number R_e doing a bad job of explaining dynamics because the distribution is highly skewed. Japan's COVID contact tracing (and now Massachusetts's also) are using *retrospective* rather than prospective tracing. In brief, when someone tests positive, the standard prospective approach is to try to track down the people they might have given it to -- their children in the transmission tree. In retrospective tracing, you instead look for their *siblings* in the transmission tree: other people who might have become infected the same time they did. In a regime where lots of transmission happens at "superspreading events", you get a lot of benefit from this change -- even after you pay the latency cost of trying to find people one infection-generation later in their disease cycle. Here's some analysis of the dynamics: Twitter thread: https://twitter.com/AdamJKucharski/status/1293472353758871552 Preprint: https://www.medrxiv.org/content/10.1101/2020.08.01.20166595v1.full.pdf --Michael -- Forewarned is worth an octopus in the bush.