(In the paper it’s slightly different, in that a context-free version is defined where a k-mer is ‘good’ if the sum of it’s \(0\pmod w\) characters is larger than the sums for the other equivalence classes, and then there is an argument that good kmers close to a UHS, and turning them into a real UHS only requires ‘few’ extra kmers.) Golan et al. (2025) Unlike UHS-based methods that optimize UHS size, this directly optimizes minimizer density by minimizing the number of charged context: Repeatedly pick the next kmer as smallest that is in the smallest fraction of charged contexts. Alanko, Biagi, and Puglisi (2025) frequency bounded minimizers, with frequency below \(t\) Prefers rare kmers as minimizers variable length scheme.