This document discusses a method for accurately reconstructing repeats from long read sequencing data. It involves computing all substrings that occur in different positions in the genome from each read, clustering these substrings into groups corresponding to distinct repeats, and modeling each repeat based on the clustered strings. Key aspects include handling different replication modes like whole, prefix, substring, and periodic repeats. The method is fast and can automatically reconstruct all repeats in a genome de novo from noisy long reads.