Random quadratic forms have been a subject of interest in the community, and the paper makes notable technical contributions appreciated by the reviewers. On the flip side MDS is a very specific form of dependence, and the presentation focuses almost exclusively on probabilistic aspects, so that the paper can appear niche to the NeurIPS public. Overall the paper is recommended for acceptance in NeurIPS, but it is strongly advised that the authors develop the application (convincingly mentioned in the rebuttal) to linear context bandits and possibly emphasize this motivation up front to increase visibility and interest in the context of NeurIPS.