Abstract
This paper discusses some design issues in the implementation of the law of uncertainty propagation according to an automatic differentiation strategy in the context of a simulation engine supporting the construction and the interactive testing of models of dynamic systems. The proposed solution propagates not only the partial derivatives, as usual in automatic differentiation, but also the input uncertainties, so to make their various modifications visible to the user of the evaluation system, and give him the opportunity to analyze the partial contributions to the standard uncertainty of the output measurands. A tool for uncertainty propagation in a general, user-oriented, computational system, instead of a software library or a dedicated system, makes uncertainty propagation transparently computable also for vector/matrix measurands, even in the case of dynamic systems, and makes uncertainty evaluation an inherent component of computational processes instead of an optional, ad hoc, addendum to them.