What This Does Not Prove

Every part of this system has an edge, and the edges are more useful than the claims.

Not understanding. Coverage and order fidelity are measurements. A peer can read every page and understand none of it.

Not a hard guarantee. This is bounded, auditable disclosure — weaker than a cryptographic guarantee, stronger than a promise, and applicable where cryptography cannot go. Saying which of those a given corpus needs is the operator’s call.

Not the hard case. On a tailnet the peer is cryptographically identified before the first byte, the corpus is the keeper’s own, and there is no adversary reconstructing anything. This is the easy case, and a testbed — not evidence about anyone else’s network.

Not correlation-proof. Several tellings released to one peer over time may reconstruct the original. The audit log makes that visible afterwards; it does not prevent it.

Not bilaterally receipted, yet. Only the keeper signs. The receiving side’s countersignature is a fork, and until that fork exists a receipt is a claim rather than a fact.

Not a question about who signs for an agent. A request can be composed and filed by an agent. The grant is a keeper act. That pushes the signature to a human on one side only, and leaves the other side open.