The basic idea
Instead of asking how often a word appears in a corpus, prevalence asks how many people know the word. The result can be shown as a percentage known.
The research dataset also reports a prevalence score based on a probit transformation. That score spreads out high-recognition words, where raw percentages otherwise bunch together.
Practical interpretation
Common words are usually safe for general readers. Uncommon words may still be fine, but they are worth noticing. Rare and very rare words often need context, substitution, or explanation.
Rarity labels are based on percent known: Common is 90% or higher, Uncommon is 75-89.9%, Rare is 50-74.9%, and Very rare is below 50%.