Harmony published a proposal on Sunday 6 September 2026 to switch off its own blockchain. Not pause it, not fork it, not hand it to a caretaker foundation: sunset the layer 1 that has run since 2019, snapshot the final state, reissue ONE as an ERC-20 on Ethereum, and point the emissions at an AI video business instead. Validators can begin shutting their nodes down from 10 September.
The stated reason is the part worth reading twice. Harmony says the threats posed by state actors and AI agents are simply too great. That is not a sentence about a broken cipher, it is a sentence about cost. Defending a chain is continuous work done by people who are paid to do it, and Harmony has concluded the work now costs more than the chain returns. Every chain has that arithmetic running quietly in the background; Harmony is unusual only in having said the number out loud.
Which makes it the fifty-third unasked half of fairness, and the first instalment where the thing that fails is not the proof. It is the place the proof lives.
TL;DR
- Harmony proposed on 6 September 2026 to sunset its layer 1, reissue ONE as an ERC-20 on Ethereum via an address snapshot, and redirect emissions to an AI video venture. Non-binding; validators may power down from 10 September.
- The reason given is economic, not cryptographic: security is not a property a chain has, it is a payroll somebody keeps meeting.
- A fairness proof is a claim about a moment. Running the chain it is checked against is a claim about the future, made by people who are allowed to quit, and no block records whether anyone still intends to.
- The migration preserves balances, not positions: multisig safes, liquidity pools and on-chain applications do not come across, and users are told to exit contracts before 10 September.
- August proved the softer half: 141,628 blocks and 109,126 transactions proposed for rollback after a 3 trillion ONE mint, justified on the basis that most were bots. Immutability was revocable in August; permanence is revocable in September.
- Satoshie’s answer is to settle short: escrow, resolution and payout in one transaction, raffles with a deployed end time, and no product that needs us to exist in 2029.
Give Harmony its due first
The mechanics are more considered than most crypto wind-downs. Snapshot holdings, delegations, rewards and contract state at a final block. Mint replacement ONE on Ethereum and airdrop it to identical addresses, so nobody has to claim anything. Supply and emission rate unchanged. Contract, snapshot calculations and airdrop scripts published for audit. Validators get a $1.372 million pool across four quarters for winding down cleanly rather than vanishing.
That is the honest exit, and it is rare. Most chains that stop being defended never announce it. They go quiet: the explorer stays up, the bridge keeps accepting deposits, and one day the RPC starts timing out. Harmony’s record is genuinely bruised, from the roughly $100 million Horizon bridge compromise in June 2022 to a staking flaw in December 2023 that minted 146.28 million unauthorised ONE. Deciding you cannot keep winning that fight, with a date attached, beats pretending for another three years.
So the objection here is narrow, and it is not fraud. Even the good version of this leaves players holding a loss no fairness proof has ever mentioned.
A proof is a moment; a chain is a payroll
The promise of provable fairness is that you need not trust an operator’s word. You verify: the VRF output was the unique output for that public key and that seed, the contract executed as deployed, the winner was selected with nobody’s thumb on it. All true, all checkable, and all of it checkable against a chain somebody is still running.
That second half never appears in the proof. Verification needs live infrastructure: nodes storing history, an endpoint serving it, clients patched against attackers who improve every year. None of that is cryptographic. It is operational, expensive, and performed by organisations with budgets, boards, and the option to redeploy their engineers into an AI video startup. Fairness is a claim about what happened; continuity is a claim about what will keep happening. The chain records the first and has no field for the second.
Balances migrate. Positions do not.
This is where the abstract argument acquires a deadline. The airdrop model is address-shaped: it reads what an address holds and mints the same amount to that address on Ethereum. That works beautifully if your value is a balance. It fails if your value is a position. A multisig safe cannot migrate, because the address on Ethereum is not your safe. A liquidity pool cannot migrate, because a pool share is a claim on a contract’s internal accounting rather than a token in your wallet. On-chain applications do not come across at all. Hence the instruction to exit your contracts before 10 September.
This is not the fifty-second instalment’s problem, where one balance honestly backed two promises. It is a different silence: a snapshot records what an address holds, never what a contract owes you. Nobody is lying. The data model simply has no column for you. And a perfectly executed migration is still a deadline you might not have read.
Permanence turned out to be a policy
Harmony has demonstrated both halves of this in under a month. On 12 August an attacker exploited a cross-shard receipt verification flaw that let valid receipts be processed more than once, forging over 3 trillion ONE across six transactions. One wallet fired 534 transfers of 5 billion ONE each in 106 seconds, 477 of which landed.
The proposed remedy removed 141,628 consecutive shard 0 blocks holding 109,126 regular transactions, justified on the basis that 95.8% of them were automated activity. Read that justification carefully, because it is load-bearing: it is a claim about how much the erased history mattered. Once “how much did it matter” becomes admissible, immutability is not a property of the ledger, it is a policy of the people running it. In August that policy edited the past. In September it proposes to end the future. Same authority, two verbs.
The gaming version
Picture a raffle on a chain like this. Ticket price a deployed constant, odds readable before the sale closed, VRF proof verified on-chain before the payout callback fired. Every question a player knows to ask, answered correctly.
Eighteen months later the chain is sunset. Your token balance lands on Ethereum exactly as promised, but the escrowed pot is a contract position, and contract positions do not migrate. So is the pending draw, the deposited bankroll, the unclaimed winnings in a claims contract. The fairness of the draw is untouched; the prize is a row in a database nobody is running any more. The VRF sentence is silent here too, and that is explicitly not a criticism of VRF. A perfectly fair draw on a chain that gets switched off is still a perfectly fair draw. It is just no longer a payout.
What Satoshie does instead
Keep every exposure short enough that it need not outlive anything. Coinflip escrows the stake, resolves it and pays out in a single transaction, so no interval exists in which your money is a loose position waiting on infrastructure. Raffles escrow the prize with a deployed end time, so a draw’s dependence on the chain is measured in the life of that draw rather than the life of the platform. VRF request and fulfilment transaction identifiers are published with results, so the artefact you would verify sits in data Base posts to Ethereum. And we run no long-dated positions: no staking lock-ups, no bankroll parked with us for a year.
Short settlement is not a proof. It is a way of needing fewer promises about the future.
Where we are exposed
Three honest limits. First, we did not choose Base’s lifespan either. An L2 sequencer is an operator with a payroll, exactly like a validator set, and while Base’s data availability story is materially better than an independent chain’s, “materially better” is not “somebody promised”. We inherit this one storey up, we do not solve it.
Second, Satoshie can stop too. We can commit to the shape of a wind-down, resolving open draws and releasing escrows; we cannot commit to existing forever, and any post implying otherwise would be doing the thing this instalment objects to.
Third, short settlement is a constraint, not a guarantee. It reduces the exposure and it rules out products we might otherwise want to build. That is a trade, and you are entitled to notice it is one.
Three questions worth asking any platform you play on
If this stopped tomorrow, what happens to an open game rather than a settled one? Everybody gets settled games right. Ask about the draw paid into and not yet resolved, and whether the answer lives in a contract or a blog post.
Is my value an address balance or a contract position? Harmony’s users have until 10 September to learn that difference the expensive way. Ask what a snapshot would record for you, and whether it is the thing you think you own.
Who is paying to defend the chain I am playing on, and what happens when that number stops working? Harmony answered this honestly and in public. Almost nobody else has been asked.
Verification tells you the game was fair. It has never told you the referee will still be in the building. Those are two separate promises, and only one of them has ever been on-chain.
Satoshie runs provably fair raffles and coinflip on Base, using Chainlink VRF for winner selection. Prizes are escrowed in the contract, odds are published before you play, and draws settle short.
📷 Photo by Alexey Demidov on Unsplash
