There is a specific distribution of names of people that is very hard to fake. Think all the different first names, last names, their spellings and common names. Generating that many realistic fake names that still match the distribution of real names is hard.
So if a distribution exists and the data scientist is aware of it and knows hed need to match it within some deviation then that is pretty trivial to do in not too much time of coding and not too much runtime. If guy doesn't have that and doesn't know hes party to a fraud then its just as easy, random match first names and last names from a limited list. There's no training scenario imaginable that would be taking into account some particular name details. Guy was probably paid near 10k per hour, which I don't know, might trip me up a bit.
When I said it's "hard" I mean for the average non-technical person, or someone who doesn't have access to a real-world dataset with millions of names for comparison.
You can easily find such datasets on the dark web, FYI.
