After much RL, the model will just learn to mock everything to get the test to pass.