Tag
This paper introduces Relational Response Fields (RRF), a theoretical framework for determining when black-box LLM responses can be reliably recovered under corruption, establishing identifiability conditions and minimax bounds that separate response consistency from truth.