By using an append-only Merkle tree, you make 'edits' by adding a new piece of data with information on what was changed in the older data. You get the benefit of also keeping a version history. I think Git works like this.

Note that Git commit graph isn't really a tree, but a directed acyclic graph, in which nodes (commits) points to actual Merkle trees (trees/blobs).

(Also, git keeps the full data, not only diffs)

Interesting. I suppose deleting sensitive data is a more difficult challenge.

Not quite ELI5, but Blahah explained how Dat handles changes below:

https://news.ycombinator.com/item?id=15541392

Exactly. Then merkle trees don't help you. They only work for static documents.

letlambda's explanation for IPFS is right. In Dat, that method is used to produce an append only log out of a merkle tree. Each time you add an entry, a leaf is added to the tree, and the new head is signed. I wrote about it here [1] and have some gifs demonstrating the process.

Merkle trees can work just fine for dynamic content. You just need to use a pubkey pointer.

1. https://beakerbrowser.com/2017/06/19/cryptographically-secur...

They certainly don't make that easier. But if the system supports a mutable link, you can have documents that always link to their own latest version.

IPFS does this by storing (mutable link, contentID) pairs in the DHT, which can be updated if you know the private key for the mutable link.