<p>Paradoxes are often treated as uniform indicators of theoretical failure or inconsistency. This paper argues that such a view obscures important differences in how paradoxes affect scientific reasoning. It proposes a diagnostic framework that evaluates paradoxes in terms of their impact on inferential viability, understood as the capacity of a framework to sustain stable and non-trivial inferential operations. On this basis, a distinction is developed between collapse-type paradoxes, which undermine the stability of core inferential processes, and persistence-type paradoxes, which generate tension between distinct inferential roles without compromising the operational coherence of the framework. The distinction is illustrated through a comparative analysis of the liar paradox and the old evidence paradox. The former exemplifies a case in which truth-theoretic inference becomes unstable, thereby motivating structural revision of the underlying framework. The latter reflects a misalignment between probabilistic updating and explanatory evaluation, yet leaves inferential practices intact. This contrast demonstrates that paradoxes differ not merely in logical form, but in the way they constrain inferential activity. The paper argues that recognizing these differences has important implications for theory choice. Collapse-type paradoxes exert revisionary pressure on theoretical frameworks, whereas persistence-type paradoxes call for conceptual differentiation or coordination among multiple inferential standards. By reframing paradox in terms of inferential viability, the analysis provides a more precise account of how paradoxes shape the limits and development of theoretical reasoning.</p>