After reading the authors feedback, the reviewers were mostly in agreement about the merits of this work (originality, potential significance) and limitations (highly engineered, not perfectly understood). Overall, this is considered as being a paper worth presenting at NeurIPS. However, It seems that authors do not appear fully committed to releasing code, which is something that the NeurIPS PC strongly encourage. This will be very useful also to get this paper read and cited by peers.