Really like this framing. A lot of agent loops today are built around the idea that every outcome is binary: either you're done or you need to try again. But in practice, there's a third state: we don't have enough information to decide yet.

That's an important distinction because failure and uncertainty aren't the same thing, and they shouldn't trigger the same behavior. Treating them differently makes the loop much more resilient. It also reinforces the idea that verification shouldn't just return a boolean. It should explain why it couldn't reach a decision so the loop can make a better next decision instead of blindly retrying.