Genome assemblies and annotations are not static and need support for tracking their evolution

For over 25 years, genomic data have been distributed in two key file formats: FASTA and GFF. These formats are used in nearly all genomic analyses and encode both genomic sequence data and the positions of annotated features. Long-read sequencing, chromatin conformation capture, and advanced assembly algorithms now enable chromosome-level assemblies and pangenomes even for the largest eukaryotic genomes. Genome consortia routinely update assemblies and re-annotate gene models as methods and knowledge improve, yet these updates occur without systematic documentation of what has been modified. As genomics enters the next era, the lack of systematic versioning becomes limiting: different annotation versions cannot be computationally compared, algorithmic improvements are invisible to downstream users, and accumulated biological knowledge exists only in human-readable documentation disconnected from the data itself. Conventional flat-file formats lack the structure to reflect this evolving landscape. While software engineering solved analogous challenges with version control decades ago, for genomics, version control is left to researchers to organise with filenames, directories, or README files. This approach cannot scale to the continuous generation and improvement of millions of genomes. We examine the limitations of genome file formats, demonstrate why incremental improvements are insufficient, and argue that genomics must adopt version control with the same gusto that is applied to generating new sequencing data. Drawing on lessons from software engineering, we outline requirements for better scientific collaboration, machine-readable formats that can capture changes, maintain complete provenance, and enable the reproducible, large-scale biology that the next 25 years demands.