The number 20% of Stockholm infected comes from modeling and testing.
I am guessing more modeling than testing, because there is very little mention of the latter.
I have posted about modeling before. I have a HUGE problem with it. Unless the modeling is based on some very extensive data from experimentation and observation, it is worthless (GIGO). There is, as yet, not an extensive body of knowledge that can be used to model the extent of Covid-19. On the other hand, the case count is kept up-to-date, and that is the only piece of data I would rely on.
The testing has been problematic. Antibody testing is subject to error, and so far, none of the antibody testing of the general population has revealed anything outside of the margin of error.