Probabilistic reliability paper

Probabilistic systems — including Large Language Models — are widely regarded as unsuitable for high-stakes environments due to non-deterministic outputs and the absence of intentional semantics. This paper challenges that conclusion by drawing a principled distinction between component-level uncertainty and system-level reliability. Using Constraint Productisation Theory (CPT v1.3), it demonstrates that reliability is not a property of the generative component — it is an emergent property of the constraint architecture governing when outputs are accepted, validated, and deployed. The paper formalises the conditions under which probabilistic systems produce decision-grade outputs and introduces constraint stability as the governing variable.