Storing each unique block of data once, however many files or snapshots reference it.
Deduplication is the technique of storing each unique block of data exactly once, no matter how many files, snapshots, or devices reference it.
The engine splits every file into blocks, hashes each block, and keys storage by that hash. If you duplicate a 4 GB Final Cut library into a second folder and back both up, the second copy costs near-zero bytes: the blocks already exist, so only the manifest entries are new. The same applies across snapshots — tomorrow’s backup of a mostly-unchanged project inherits yesterday’s blocks by reference. Deduplication is what lets a repository hold a year of daily snapshots for a working drive in a small multiple of the drive’s live size.
It also works across backup sets on the same machine, and on some configurations across devices pointing at the same repository. A font file sitting in ~/Library/Fonts and bundled inside a Sketch document resolves to the same block on disk. So does the identical row of video frames that survived a timeline trim.
In macup, deduplication is why your first full backup is slow and every subsequent one is fast. The initial run hashes and uploads everything; from then on, macup only moves blocks it has never seen before. Restoring a 40 GB project from a six-month-old snapshot pulls a mix of blocks written over those six months — the engine stitches them back into the exact files you had.