Abstract: 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). Comparative testing across contemporary LLMs confirms that most models obey SISO in mathematically predictable ways. Except GROK. That one belongs in the gutter.
