- avoid data corruption when downloading/transferring/copying datasets

- notice changes/updates in the original dataset

- dataset versioning (think how e.g. git turns directories and files into hash trees -- also called content-addressing)

- most importantly: stable names without a naming authority