Why waste a good polynomial

In the past week Jan finished the implementation of Taylor series methods for ODE-IVPs. The algorithm uses forward-mode automatic differentiation to recursively compute Taylor coefficients for the field and solution components of increasing length and the implementation relies on memorization of sub-expressions to guarantee that syntactically equivalent expressions only get computed once. Jan has been reflecting on the fact that the algorithm actually produces the Taylor polynomial for the field and solution components, but then collapses them into a single interval box. While the current line of inquiry aims for an algorithm where components exchange approximations given by intervals, and not polynomials, Jan hopes to return to investigate the potential of keeping the polynomials unevaluated, and applying the algorithm on them directly, as a potential means of combating wrapping effects. Jan wonders if this is one of the techniques employed by advanced solvers such as VNODE or ValEncIA-IVP, which also rely on Taylor series methods and conditioning techniques similar to those proposed by Lohner and Zgliczynski.