Active centromeres share a conserved array-architecture signature across great ape genomes

Complete telomere-to-telomere assemblies now resolve centromeric repeat architecture at nucleotide resolution, but identifying which of several candidate α-satellite arrays is functionally active still usually requires CENP-A or methylation data. Here we ask whether active centromeric arrays occupy a recurrent architectural state that can be recognized from assembled sequence alone and whether that state is conserved across great apes. Using sample-matched CENP-A occupancy in T2T-CHM13, we find that active arrays are both expanded and more sequence-homogeneous than neighbouring relict arrays. The longest candidate array identifies the functional centromere on all 19 testable human chromosomes, whereas a homogeneity-based model we created, Centroseek, identifies 17 of 19. Functional assignments were robust across independent CENP-A datasets. Across 47 eligible chimpanzee, gorilla and bonobo chromosomes, the longest-array rule identifies 44 of 47 annotated active arrays. All three failures occur in bonobo, where relatively small active arrays break the otherwise strong expansion pattern. Notably, CentroSeek recovers these three bonobo exceptions, although it does not outperform array length globally (38/47 versus 44/47). Array length and homogeneity are strongly coupled, and homogeneity provides little additional predictive information once size is known. Together, these results define a conserved expansion-and-homogenization architecture of active α-satellite arrays, with expansion providing the dominant sequence-only signal and CentroSeek highlighting exceptional centromere architectures in which size alone is insufficient.