Figure 1b confuses me: in my experience, (double) decycling is never worse than random minimizers, while in the figure decycling sometimes is worse. Since these minimizers can be selected in streaming with minimal computational overhead: the original decycling set paper (Pellow et al. 2023) does not provide code for streaming computation of the minimizers. Since non-lexicographic minimizers are used, there should be little/no correlation between bucket sizes of lexicographically-close minimizers.