Reachability analysis — supplies the safe set that bounds → Robust policy improvement

These two are cited together three separate times as a pair of tools for building conservative baseline policies (concrete-problems, §"Define an Impact Regularizer:", p. 5; §"How to respond when out-of-distribution.", p. 18), and the repetition reflects a genuine division of labor rather than redundancy. Reachability analysis answers a set-membership question: from a given policy, which states can the system reach, or safely return to? It defines the boundary of what counts as safe. Robust policy improvement answers an optimization question: given uncertainty about the true dynamics model, how should a policy be updated while still guaranteeing a performance floor? It operates inside whatever boundary reachability analysis has drawn. Neither alone produces a usable conservative baseline: a reachable-set computation without a policy-improvement method has nothing to act on, and a robustness guarantee without a defined safe region has nothing to guarantee performance within.