Comment by mcguire

That's what I was wondering. If so, submitting many test runs makes it trivial to provide a submission that works perfectly against the test data specifically.