I propose there exists a confusion-recovery mode for conversation. It triggers when what you just heard makes no sense to you. Evolution expects this to happen when the speaker is rationalizing or dishonest. The instinct is to get suspicious, stop interpreting words literally, and look for an in-framework explanation that fits the other person's expected bias. I also think this has an element of preferring premature certainty to valid confusion.

If there's a huge inferential gap, you'll trigger this mode immediately.