Cursor uses Merkle trees - for any change, it takes O(log(n)) bandwidth to find where the client/server disagree. Like bloom filters, a Merkle Tree is a data structure where every leaf node stores the hash of a block of data, and every parent stores the hash of its children. > Every file gets a hash, based on its contents. The leaves of the tree are files. > Every folder gets a hash, based on its hash of its children