I have taken several courses in statistics and am using R (with the survey package) for my analysis of ACS data on a person-level just for the year 2017. As I understand the person weights, a manual duplicate approach to weights should be exactly the same as using a statistical package to incorporate weights (given that no manual errors are made, and of course it would make calculations slower because of increased size of the data). After reading this post ( here: Person Weight), I was concerned that I don’t have the right understanding of the person weights—can you confirm?

Yes, you are thinking about this correctly. In IPUMS USA, the PERWT value indicates how many persons in the US population are represented by a given observation in your sample. That means if you correctly perform a manual approach your population counts should be approximately equal to the use of a statistics software package. As you can see from the answers to the post linked above, some researchers hold a different perspective on this topic. They make the case that a more accurate method for calculating standard errors around point estimates or population counts need to take into consideration additional factors of the complex sampling design. For this approach the variables CLUSTER and STRATA are helpful. The use of sampling weights is an often debated topic among applied researchers, so that is why you may find conflicting information about how to handle sampling weights.