You could just hard code a PRNG seed, and then the test would trivially be deterministic? I'm not sure what they say their objection is, is actually their real objection?