Merkle Proofs#

Every atom stored in Parametric Memory is inserted into a Merkle tree (RFC 6962 Certificate Transparency spec). When you recall an atom, you receive not just its value but a Merkle proof — enough data to independently verify that the atom is part of the committed tree, without trusting the server.

Why this matters#

Standard databases require you to trust the server. If a memory system returns a fact, you have no way to know if it was tampered with since it was stored. Merkle proofs change this:

Store:  hash(atom) → inserted into tree → new rootHash published
Recall: atom + sibling_hashes → client recomputes root → compare to known root

If the computed root matches the published root, the atom is authentic.

Proof structure#

interface MerkleProof {
  leafHash: string; // hash(key + value + version)
  siblings: string[]; // sibling hashes from leaf to root
  rootHash: string; // tree root at time of proof generation
  leafIndex: number; // position of this atom in the tree
  treeSize: number; // number of leaves at proof generation time
}

Verifying a proof#

The SDK exposes a verify helper that runs the proof locally:

import { memory } from "./lib/memory";
 
const { atom, proof } = await memory.recall("v1.user.Glen.preference.theme");
const valid = await memory.verify(atom, proof);
 
if (!valid) {
  throw new Error("Memory tampered — proof invalid");
}

You can also verify raw proofs without the SDK using the verify-merkle-proof utility included in this repository at src/lib/verify-merkle-proof.ts.

Consistency proofs#

For auditing, you can request a consistency proof that proves the current tree is a superset of a previous tree — nothing was deleted:

const proof = await memory.consistencyProof({
  fromSize: 500, // tree size at an earlier checkpoint
  toSize: 1200, // current tree size
});

Consistency proofs are only available on Professional and Team tiers.