Normative language for conformance testing requirement

Following on from the discussion on yesterday's call, the two forms
of language suggested for a normative requirement for conformance
testing were along the lines of
  1) A conforming implementation MUST be able to pass all the tests....
  2) A conforming implementation MUST NOT fail any of the tests....

On the call it was stated that these are logically equivalent.

My preference is for the second form.  An advantage of this form
is that the statement is clear cut and easy to verify.  To show that
an implementation is non-conforming, it is only necessary to produce
a single test that fails to run with that implementation.

The language of the first form with its use of "be able to" seems
less clear and harder to verify.  The "be able to" language means that
there is no requirement to actually run any of the tests.  Without
doing this, how could it be shown that an implementation conforms to
the testing requirement (or not)?

   Simon