What do you mean by it "should be easy"? It isn't easy at all; nested inductive types is the example that come to mind. Something like HOL is easy to check by hand as a prover kernel but Lean's type theory is definitely not.