Of course it's possible that the test is so divorced from reality that you can substantially increase real-world emissions without affecting the test, and that would be a bad test.
But if the test doesn't allow you to predict actual emissions, as long as it roughly identifies "more" and "less" then it can work for a regulatory lever.
But if the test doesn't allow you to predict actual emissions, as long as it roughly identifies "more" and "less" then it can work for a regulatory lever.