I'm using it for SQL queries. My query returns subset of some column set. It's not practical to define a separate class for every query. So all queries for the given table returns the same class representing all columns in a given table. But if one method returns subset of those columns, other columns contain uninitialized value and should not be read. Any attempt to use uninitialized value is a bug and must be caught as soon as possible.
I'm not sure it's really impractical, at least in my experience. You may not need a separate type for every query, as long as you are willing to have some queries select a few more columns than you actually need right now; and you don't need separate classes, just separate interfaces, which makes the amount of extra code you need smaller.
Oh, the wonders of using widespread mutable state! Did you remember to call this other method here before trying to call me??? No?!?! Oh, here's a nasty Exception for you then.