In the Hashable ABC, there is a peculiar code fragment:

                if "__hash__" in B.__dict__:
                    if B.__dict__["__hash__"]:
                        return True
                    break

When would the innermost if-statement ever be False?  Is there a reason 
to define __hash__ to be something that evaluates to False?

This is precisely under discussion in issue2235:
if a base class is hashable, a derived class may set __hash__ to None,
and disallow hashing.

I don't think we should build in explicit support for bad designs that 
violate the Liskov substitution principle. Are there any valid use 
cases for wanting non-hashable subclasses of hashable classes?

If for some reason, this feature survives, it would be better to use 
NotImplemented instead of None.

There are tons of ways to violate Liskov in Python. Liskov is not always
the right rule.

NotImplemented is not appropriate -- it is only used as a magic return
value from binary functions.