The data here is processed from NHTSA's FARS database. When someone dies in an accident, it gets input into a STATE reporting system, and FARS is manufactured by analyzing each state's individual record system. The feds consolidate all this data and publish a unified dataset annually. They say it's "a lot cheaper and just as good as collecting it themselves"

Additional errors are potentially produced from my own processing of the federal data, but those will be rooted out over time. Project being OS will hopefully help with that.