arXiv:2611.13986v1 Announce Type: new Abstract: Stochastic model predictive control uses disturbance statistics to set safety margins. Errors in these statistics can invalidate the safety guarantees. This paper therefore describes the uncertainty in the disturbance mean and covariance by a prescribed Gelbrich ambiguity set. This contains distributions of any shape whose moments are close in the Gelbrich distance to the nominal values, including distributions with unbounded support. For linear time-invariant systems with additive disturbances, we build on the indirect feedback paradigm, propagate this moment uncertainty along the error dynamics, and obtain a closed-form constraint tightening that is computed offline. Robustness enters as a single additive term that grows with the size of the ambiguity set, so the online problem remains a quadratic program. Recursive feasibility and closed-loop satisfaction of conditional value-at-risk (CVaR) constraints hold for every disturbance distribution in the set. We illustrate the method on a linearized planar quadrotor.