Classical accuracy survives the exponential lift. An Evidence Press briefing. Many equations describe rotations or other constrained matrices. Ordinary numerical updates can leave that constraint, while exponential updates preserve it. But matrix multiplication does not commute, so preserving the shape does not automatically preserve accuracy. This paper proves that a structured family is different. Take a Williamson low-storage Runge–Kutta method, with any finite number of stages. If its classical order is at most six, the specified exponential version has at least the same order. The argument allows zero and negative coefficients and repeated stage times. The proof starts with an identity for the ordered products produced by the updates. Small exact linear systems then show that the extra conditions follow from the familiar classical conditions. A continuous integration comparison identifies their exact target values. The package contains the written argument and exact checks of 197 ordered-tree coefficients. Separate internal implementations check prefixes and ranks, while deliberate corruption tests confirm that invalid certificates are rejected. A further result gives three explicit extra tests sufficient for order seven. It does not show that those tests follow automatically, and it does not settle the all-orders question. Nor does the paper promise better error constants or computational performance for a particular method. This is an anonymous, unrefereed mathematical candidate, not externally reproduced or formally verified. Published on 20 October 2026, its paper and evidence are linked on the Williamson order-transfer release at Evidence Press. This briefing uses an AI-generated voice and is not additional scientific evidence.