Abstract:This paper introduces the Compression Decay Comprehension Test (CDCT), an information-theoretic framework for quantifying model comprehension through semantic robustness under compression. CDCT measures how language models preserve conceptual integrity when information density is systematically reduced, revealing nonlinear comprehension decay patterns independent of model scale. The framework provides a reproducible benchmark for identifying reasoning-aligned architectures, differentiating genuine comprehension from statistical mimicry. Notes:This version is the author’s original manuscript released for open access and citation. An interactive dashboard summarizing the key metrics and trends from these experiments is available here: https://cdct-web-ranking.onrender.com/.