Is 'testability' a useful architectural constraint?

Having examined simplicity, it strikes me that testability is another
desirable feature. One thing that the QA activity has shown is that
good specs have testable assertions, and conformance criteria, and
test suites. This makes implementation reports and maturity assesment
much easier, and if the working group is involved in the creation,
discussion and bugfixing of the test suite then the overal quality of
the spec also seems to benefit.

If a spec is not testable and conformance cannot be assesed, it makes
it hard for other specs to interact with it in consistent and useful
ways. It can also show that the working group thinks it has consensus
on what the spec does, but is wrong ;-)

Thus, testability aids clean architecture.

-- 
 Chris                          mailto:chris@w3.org