The axiom “shit in, shit out” (SISO) has governed system design for decades. We introduce a structured evaluation framework for Large Language Models (LLMs) composed of two diagnostic instruments: F.A.R.T. (Factual Ambiguity & Relevance Test) and T.U.R.D. (Token Utility Response Dependability).