The contribution is well-motivated, resulting in a simple method that is easy to implement, and providing a well-calibrated estimate of the uncertainty in the reported experiments. The novelty is rather moderate, but the developments are rigorous, pointing towards approximations where necessary. Please make sure to release code.