← Meridian · Notes

Meridian update · 10 Oct 2026 · claim-labeled · hypotheses until tested

Research direction: Nightframe comparison and the revocation-in-flight question

Meridian remains a core Society of Explorers project. We are examining the new Nightframe claims, with Cardano’s eUTXO model and Midnight’s privacy tools as supporting infrastructure, against one practical question: what may an AI agent still do when its permission is withdrawn but the cancel has not arrived? We will publish sources, comparisons, small experiments and objections as we go. The research is open to scrutiny, and the technical claims remain hypotheses until tested.

current hypothesis open

The question

open An agent is allowed to spend, order, or move something on someone’s behalf. Its permission is withdrawn. The withdrawal has not reached the agent yet, or it has reached the agent and the agent has not acknowledged it. What may the agent still do in that window, and who answers for what it does? This is the same gap described in The cancel has not arrived, narrowed to delegated spending and a connection that can drop.

Why Nightframe, and what we are not assuming

announced Nightframe was introduced by Charles Hoskinson, Cardano’s founder, in public talks, most fully in his Midnight City Tokyo keynote on 10 Oct 2026. He describes it as a Midnight-based framework that combines zero-knowledge proofs, multi-party computation and trusted execution. We treat this as an announcement. We found no released specification or code, and we keep what has been announced separate from what has been shipped.

The offline-payment claim, in his words

announced In the Tokyo keynote (14:56–16:40), Hoskinson describes paying offline. Alice’s money “lives in her phone” in “one of these trusted computers,” which holds “a unique private key.” She taps Bob’s phone and sends “a copy of that unique private key.” Once Bob confirms he has it, her copy is erased, and she can “send over a proof of secure erasure” (probable wording; YouTube’s captions read “eraser”). He calls this offline, off-chain and “quantum money.”

hypothesis As described, this is a classical design. Its safety rests on trusting the phone’s secure hardware to erase the key, and on whoever vouches for that hardware. It is not quantum money in the scientific sense, which relies on quantum states that cannot be copied (no-cloning). Ordinary bits, a private key included, can always be copied. The talk gives no timeline, and we found no released specification or code. It does not say how copied keys, backups, the recipient’s checks or settlement would work. Those are the questions below.

current Midnight is the supporting infrastructure in this comparison. Its own update reports that permissionless contract deployment is live on mainnet. Cardano’s extended UTXO model is the supporting design for one narrow property: a spending condition that is explicit and can be used once.

Three things stay separate on this page. A single-use output is not offline finality. Selective disclosure states only what a given proof establishes. Post-quantum cryptography, which is a set of published standards, is not quantum money, which depends on quantum states that cannot be copied.

Test order

1. Ordinary baseline first. A signed record plus a controller that issues expiring permission and enforces it locally. Every later option has to beat this, not a straw man.
2. Cardano eUTXO only for explicit, single-use spending conditions.
3. Midnight only where selective disclosure is the documented requirement.

The experiment logic stays portable. A chain gets chosen only when a written requirement and a matched comparison justify it. No new chain is being started.

What any offline-spending answer has to cover

• Copied keys and wallet backups.
• Rollback to an earlier state.
• The same value spent twice while disconnected.
• How the recipient verifies what it is accepting.
• Settlement and reconciliation after reconnecting.
• What it relies on: secure hardware, a trusted issuer, a bounded amount at risk, or later reconciliation.

hypothesis For a disconnected machine, the working assumption is that permission has to expire on its own and be enforced locally, because a remote cancel cannot be presumed delivered.

What this update does not claim

• No partnership with Cardano, Midnight, or any company behind them.
• No claim that Meridian is novel, has a unique primitive, or is audited as a protocol.
• No claim that offline payment or quantum money is released or working.
• No reopening of claims the September 27 audit closed. The original G2 commutativity claim stays rejected. What the audit left unsupported still stands.
• No token, no real funds, no hardware purchase.

REB v0c: two reporting corrections

Presentation errata · 10 Oct 2026 · preliminary comparative simulation

The REB v0c comparative simulation and its independent review are complete. The review supports a preliminary simulation note with the two corrections below. This presentation update leaves the frozen source, results and decision rules unchanged. The no-advantage conclusion stands: there is no demonstrated Meridian mechanism, advantage, production security guarantee or buyer demand.

1. Budget overdraft covers the whole run

The package’s results note described whole-run budget overdraft as “pre-revoke overspend” and placed it outside post-withdrawal exposure. The calculation is actually max(0, total committed − 100), with no pre-revoke filter. Among configurations meeting the held-out post-withdrawal exposure criterion, excluding sensitivity runs, the largest observed whole-run overdraft is 32 units.

In the archived S2 example, 123 units were committed before withdrawal and 9 after: 132 against a budget of 100. Of its 32-unit overdraft, 23 already existed before withdrawal and 9 arose after. In the S3S example, 88 before and 19 after total 107, so its 7-unit overdraft arose after withdrawal. Post-withdrawal commitments may contribute to both measures. Exposure feasibility and shared-budget conservation are separate properties. These simulation results are not evidence of a Stripe deployment flaw.

2. Outstanding holds bound a local settlement tail

The package’s builder question described preserving in-flight holds as bounding post-revoke exposure by outstanding holds alone. Under the stated hold rules, that bound applies to the settlement tail after a particular enforcing authority has observed revocation and stopped new reservations.

It does not bound all exposure measured from the owner’s withdrawal issuance. Other authorities may still accept new commitments before learning of the revoke. Keep owner issuance and each authority’s observation time, and the amounts committed in those intervals, separate. Grant expiry, reservation or hold expiry, and revocation also need separate rules; grant expiry alone must not be assumed to cancel an existing reservation or commitment.

Source: the completed REB v0c source and results package, read alongside the newer independent review dated 10 Oct 2026. These are reporting corrections to the package’s results note and builder question; the experiment was not retuned or rerun for this update.

Next

The next useful evidence is a builder’s implementation or specification decision: assess the completed REB v0c evidence against that implementation’s enforcement, revocation, grant and reservation assumptions, then identify any further evidence required before development.

Sources

• Charles Hoskinson, “What comes after the smart contract? Welcome to Nightframes”, recording posted on Charles Hoskinson’s own YouTube channel, Midnight City Tokyo, 10 Oct 2026. Primary source for the Nightframe claims; offline payment at 14:56–16:40. Quotes are checked against YouTube’s captions, three local machine transcripts and the on-screen slides. None of these is a human-verified transcript, and where the machine readings differ we say so.
• Aaronson et al., “Quantum Money,” Communications of the ACM, 2012
• Midnight: State of the Network, September 2026 (first-party)
• Cardano: extended UTXO model (first-party)
• NIST: post-quantum cryptography
• The cancel has not arrived · Research evidence and objections · Meridian