tantekjust confirmed btw that the CSS WG does write tests for SHOULDs in their specs, but then document the test accordingly in itself, that it is testing a SHOULD
tantekif it turns out that implementations all (or nearly all) pass a test of a SHOULD, that makes a good case for making it a MUST in an iteration of the spec
tantekif the reason for a SHOULD is "some implementations may not want / need to" then create an implementation profile / conformance class accordingly that either includes those requirements as MUST or not