Error and NativeError can both be used to initialise any error subclass instances, because both initialise the [[ErrorData]] internal data property. But the current text says Error must be used to initialise Error subclasses whereas NativeError must be used to initialise NativeError subclasses.
test case:
---
js> class MyError extends Error {constructor(){}}
js> err = new MyError(); err === TypeError.call(err)
true
js> class MyTypeError extends TypeError {constructor(){}}
js> err = new MyTypeError(); err === Error.call(err)
true
---
fixed in rev17 editor's draft
I changed a could "musts" to "shoulds"
I think this overall behavior is probably ok. However, it may not be valid if am implementation attaches additional state to native errors (eg, a stack trace) that is not also associated with generic Error objects
fixed in rev17, August 23, 2013 draft