Jev judgments could help search AI chats by frustration or satisfaction

A user describes building an “embedding” of agent traces with Jev judgments: a set of numeric scores where every dimension has a name. Their example tracks frustration, follow-ups, whether the user corrected the agent, and satisfaction. They say these scores let people cluster sessions or search for “frustrated users who had to correct the agent” while seeing why each result matched. The same approach, they note, can find happy users to help identify what is going right.