> encouraging the use of a standalone signature for type declarations where at least one parameter of the datatype does not have kind Type.
So Dict, Eq both get a sig but Fix and Either do not?
type Dict :: Constraint -> Type
type Eq :: Type -> Constraint
type Fix :: (Type -> Type) -> Type
That's not how I understand Richard's criteria. Dict and Fix have non-Type parameters (Dict has a Constraint parameter, and Fix has a (Type -> Type) parameter. On the other hand, Eq and Either have only Types as parameters. This seems to match my intuition about when a kind signature might be helpful, as well as yours as far as I can tell from what you wrote.
That's not to say I am advocating any kind of rule. As I'm not really involved in GHC development, I refrain from having any opinion. I just think you may have misread Richard's suggestion.