I found it not just to lead to embellishment, but (1) the problems it did flag mostly would be caught by minimal testing; whereas (2) it regularly missed deeper problems. For an example of the latter: using TanStack (React Query) api caching, you have different data shapes for infinite scroll vs non infinite scroll. There were circumstances were an app confused them. Typescript had nothing to say. Nominal typing easily handles these cases and, ime, caught more actual problems.