Kinda trivial but...

Something like the enclosed, untested patch would seem to make new_class work a bit more like type_new.

To be explicit, the difference is whether or not to respect virtual subclasses.

So, for example, as implemented, we can implement __subclasscheck__ on a 'AtypicalMeta' metaclass to create an 'Atypical' class whose __mro__ is (ATypical, object) -- note, no 'type' -- but for which issubclass(type, ATypical) is true.

If I'm not mistaken, this disguise will suffice sneak our psuedo-metatype past new_class (but not type.__new__, so we still get the same error message, and the universe does not implode on itself as planned...)

In my case,the only sequela was that the fantasy of my very own type-orthogonal graph of "foo"-style classes in 3.x was first subtly encouraged and then dashed against the Cpython rocks... (just kidding, sort-of).