Trusted policy oversight — operationalizes the recoverability check in → Bounded exploration

Bounded exploration states the general requirement abstractly: an agent should be confined to a region of state space where "even the worst action... can be recovered from or bounded in harm," checked, if a model is available, by extrapolating forward to see whether a candidate action would exit that region (concrete-problems, §"Bounded Exploration:", p. 15). Trusted policy oversight supplies one concrete way to run that check: instead of an agent verifying recoverability against its own model, a separate trusted policy paired with an environment model makes the call, and exploration is limited to actions that trusted policy believes are recoverable (concrete-problems, §"Trusted Policy Oversight:", p. 15). The difference is who does the checking — trusted policy oversight delegates the safety judgment to a second, presumably more conservative, policy rather than folding it into the exploring agent itself.