I think you are right. I was expecting something different, but yeah, this actually makes sense. Instead of translate the content of the string table, the localization tool will actually get the texts from the FTexts being used in the project, similar to when we are not using string tables. I know this localization workflow may have some advantages (like keep tracking for changes or updates), but I was expecting to be able to only work with .csv files, one for each language.
And again, it would be very helpful if the documentation was more explicit about how it work.