Despite the lack of scalability of the proposed approach, this paper brings interesting new theoretical insights on the structured bandit problem. More comments on the complexity of the CROP algorithm and on whether doing experiments is actually possible would be appreciated.