I also believe this to be true, but it is more complex than just taking the last 50 years. What FIREcalc does is, using the set of actual economic data running from 1871 to 2020, including inflation for the year, run as many simulations as your specified time horizon will allow, using all the information you input regarding portfolio amount and composition, spending, pension income, social security etc. So, for example, if you specify 50 years, it starts with the assumption that you are retiring in 1871. It applies the 1871 returns for each asset class, the inflation for 1871 (to determine what your 1872 spending will be), deducts your spending, etc and sees where you are in 1872. Then it applies all the factors from 1872 and so on and so forth until you have gone 50 years, ending in 1920. That is one run. Then it starts again only assuming you retire in 1872 and goes for 50 years, ending in 1921. If keeps doing this until it runs out of data, which means it will run 70 times, the last run will assume retirement starting in 1971 and ending in 2020. Then it determines how many of those runs fail -- that is, where you run out of money before the end. If 7 runs fail out of 70 (10% failure) it will tell you that you have a 90% success rate.
In sum, it applies actual inflation data from the past. It does not anticipate future inflation. If you just plug in 3% inflation, that will be higher or lower than any particular year really was (very unlikely to have been exactly 3% in any year).
Edit to Add: I don't know what inflation measure is used from 1871 to 1912, because I think CPI only goes back to 1913