And it's specifically the low error rate of the HiFi sequencing that enabled this advance. The string graph collapses not just because of sequence length, but error rate that makes it impossible to distinguish repeats from each other. The 15kb reads used in this paper have just enough length and accuracy to span almost all repeats in the genome (rDNA and some centromeres excluded).

The assembly used the same conceptual foundation as the original human genome assemblies, for the curious: https://doi.org/10.1093/bioinformatics/bti1114