One subtle thing Walrus gets right is what it actually proves.
Walrus doesn’t just care that a blob exists somewhere on disk. It cares that the blob can be retrieved when someone asks for it. That’s why the protocol uses proofs of availability, not vague proof that data was once stored.
For every active blob, assigned nodes must regularly demonstrate that they can serve their portion of the data. These checks happen continuously across epochs. If a node stops responding or cannot produce its piece, rewards tied to that blob stop.
This distinction matters for applications. A stored blob that can’t be fetched is useless to a game, an AI pipeline, or an archive. Walrus treats retrievability as the product, not storage claims.
I think of it like a library. It’s not enough that the book is “owned” by the building. You need to know it’s actually on the shelf when you show up.
Walrus prices and rewards availability, not promises.