Right, that's why it seems that the only reasonable way is to look at all cause mortality. That's the only number that can't be manipulated (but can have time delay when they are recording data). Everything else is subject to manipulation (what is definition of COVID death,with COVID or due to COVID, what is a positive case, etc.).

I also clearly remember that in April/May 2020, lots of people were speculating that many more people were infected than official data was showing. Many countries did seroprevalence tests in May 2020. For example, Spain found about 5% (source: https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)31483-5/fulltext). But a similar Spanish study (April-June 2020) showed 10% seroprevalence among health care workers (source: https://pubmed.ncbi.nlm.nih.gov/33434269/)

And a metastudy of seroprevalence (from Jan to Aug 2020) that "The estimated number of infections, however, were much greater than the number of reported cases and deaths in almost all locations." (source: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0248946) This was a year ago. And during the times when "experts" were speculating that we are ah so close to herd immunity (estimated at the time around 20%).

So, if we look at official data (infections via PCR test -- which is tricky business as you point out) now, many countries have had 10% of population infected (US about 13.7% and the UK about 12.3%). So, maybe the real number of infections is not so tiny and may be 1.5X-2X of the official count, so 15%-20%?