Bruteforce, generate as many as possible.

Test against real curve.
Just like I normally would crack passwords.

Just thinking to reverse the process, as it is the same question for testing the statistical significance as JCL describes in the blackbook.
Running patterns against a random curve, it is the same basically, just the other way round.

Actually it is not completly random, since the random of SHUFFLE I assume is also limited, since it randomizes the previously known curve.
It is therefore only partly random.