This idea is called group testing [0] and has been used by the army for quite a while. It is also a well-studied problem in information theory as a (useful) bas

This idea is called group testing [0] and has been used by the army for quite a while. It is also a well-studied problem in information theory as a (useful) basic model and we know optimal solutions to it. What is surprising is that group testing tells us that we can actually do better: the specific construction you pinpoint the people who come up positive, without needing to have a test for each person [1] and this result is optimal in the expected number of tests. I'm surprised that most labs haven't already been doing this (but I would suspect some have). Wootters has wonderful notes on the mathematical background for this—I really recommend them. [2] ----- [0] https://en.wikipedia.org/wiki/Group_testing https://en.wikipedia.org/wiki/Group_testing [1] Of course, we require an upper bound on the total prevalence, but even a weak upper bound yields fairly good savings in testing. [2] Page 3 on of http://web.stanford.edu/~marykw/classes/CS250_W18/lectureNotes/Lecture8-compressed.pdf http://web.stanford.edu/~marykw/classes/CS250_W18/lectureNot...