Sanity Checking Causal Representation Learning on a Simple Real-World System