The discrepancies between the data sets are really strange. I downloaded a third (still older) version of the data set to check. There are some differences between the two older data sets - within reason I would say. The new dataset seems to have been constructed by some different methodology - as the "person-years" are now completely different, while the deaths also have moved around quite a bit. Looking at April 2021, the discrepancy in the "unvaccinated" deaths is about 300.

But this is not the biggest problem - no, the biggest issue is with the statistics. Take the same April 2021, All Causes as an example:
- there are 3'778 deaths in the "unvaccinated" group
- there are 27'856 deaths in the "ever vaccinated" group
- the person years are quite similar in both groups - within a factor of two

So if I calculate the death rates over 100'000 by hand I get:
- 281 deaths per 100k in the "unvaccinated" group (in the file - 2266.5)
- 1478 deaths per 100k in the "ever vaccinated" group (in the file - 862.4)

The same pattern holds true for all files, up to and including Dec. 2022:
- "unvaccinated" my calculation is 206 vs 1026.7 in the file
- "ever vaccinated" my calculation is 1'113 vs 944.9

So what's the conclusion? Apparently, I don't have the first clue what is going on with this data! I don't think the "person-years" column means what we think it means. If it did, then you should be able to estimate the population size by taking the total "person-years" and multiplying by 12, right? However by that logic the older files point to a total population around 39 million (18+) while the newer point would indicate around 47 million (18+). Both of these numbers are very far from the ~54 million I calculate from the graph in Statista.

Unfortunately I am unable to find a detailed explanation of the methodology used to aggregate this data. I must say that I am now much more confused than before.