Every transfer, drawn as it happens
Not a projection and not a pitch deck. Every rate below is the number set on Monad right now, and every arrow is a transfer the contracts actually make. Scroll, and the diagram lights up the part being described.
Before any of the numbers
Gaming rigs asleep at 3am. Workstations doing nothing all weekend. Small datacenters between contracts. Mining hardware with no coin worth mining. The capacity to run the current wave of AI is already bought and paid for — it is just scattered across millions of machines that have no way to sell an hour of themselves to a stranger.
You are told the machine is an H100. You cannot check. You cannot check whether three other tenants are on it, whether it is throttled, whether it is the card you are paying for, or whether it exists at all. You are buying a claim, and the only recourse is a support ticket to the company that made the claim.
Take the provider's word, wrap it in a token, and hope reputation sorts it out. Reputation is a lagging indicator: it tells you who was honest before they had a reason not to be. The bottleneck was never hardware. It is verification.
Under a timed challenge it cannot precompute, watched by independent parties who each sign the same statement or none of them do. Then make it keep proving it, commit memory to your job that you can challenge at any moment, and bond it so failing costs more than cheating earns.
Every number on this page exists to serve one sentence: make it cheaper to be honest than to lie, and then get out of the way of the two people actually trading.
Honestly — the market works without one. Jobs settle in USDC and always will; a provider is paid in money, not in a promise. $WRAITH exists for exactly two jobs neither of which settlement can do.
Supply will not show up before demand and demand will not show up before supply. Someone has to be paid to be early. Emissions are that payment, and they are designed to matter less as real revenue arrives — not more.
The people who put hardware on the grid should end up owning the grid, rather than renting their capacity to somebody who does. Emissions are how that happens without anyone being sold anything.
And what it must never be: the reason a machine is here. If emissions are the only thing paying, the grid empties the day they stop. That is why the larger of the two rates is paid for settled work, and why the real money — the yield on locked deposits — only reaches machines that actually did a job.
A renter pays for compute. That payment splits four ways and nothing else in the system can touch it. Everything after this chapter is a detail of one of those four arrows.
A listing carries a per-second rate, a minimum and a maximum term. Opening a job escrows the whole amount immediately, so a provider is never asked to work on a promise.
Three things are skimmed before a provider is paid, and every one is capped in the contract rather than in a policy document.
| Skim | Now | Hard cap |
|---|---|---|
| Protocol fee | 2.5% | 10% |
| Referral | 1.5% | 5% |
| Licences | ≤ 25% | 30% |
Minting a soul locks a refundable MON deposit, sized to what the machine can earn. It is staked into shMON for as long as the machine is enrolled, and the yield it throws off is split.
The other 30% funds the buyback. This is real money, not minted supply — and it only reaches a machine for periods in which it did a job. Being present earns $WRAITH; being useful earns MON.
$WRAITH is priced in WCU-hours, not in the job's currency — one scale across NVIDIA, AMD, Apple silicon, CPUs and browser tabs, where 1.00 WCU is roughly an RTX 4090.
Staking yield accrues on every locked deposit whether the grid is busy or idle, and licence sales are capacity bought forward. Neither needs anyone to rent anything.
That is what lets a buyback start before utilisation exists — and it is why the treasury is a settable address rather than a hardcoded one.
Compute demand pays providers and the treasury. Deposits earn yield that pays workers and the treasury. The treasury buys $WRAITH. $WRAITH pays providers to be there and renters to show up.
Follow one job all the way down
A renter takes an H100 for ten hours at $2.00/hour. Every figure below is computed from the parameters live on Monad right now — not an illustration.
The machine that did the work. Always the majority — the caps are written so it cannot be otherwise.
Whoever bought a slice of this machine. This provider sold 10%; 25% is the ceiling.
To the treasury, which funds the buyback. Hard-capped at 10% in the contract.
To the licence that sourced the job, if one did. Nobody sourced it, nobody is paid it.
| What | Asset | This job | Paid for |
|---|---|---|---|
| Settled-work emission | $WRAITH | 287.5 | 5 per WCU-hour, times 5.75 WCU, times 10h. A refund cuts it pro-rata; a fully refunded job earns none. |
| Availability emission | $WRAITH | 57.5 | 1 per WCU-hour online and listed. It was available as well as working. |
| Deposit yield | MON | a share | 115 MON locked and staked in shMON. 70% of what it earns goes to machines that completed work, and this one did. |
Three balances, three assets, one machine — claimed separately and not interchangeable:
withdraw for the job payment, claim-rewards for $WRAITH,
claim-yield for MON.
Why it compounds
A flywheel that only spins once demand arrives is not a flywheel, it is a hope. Two of these turn on locked capital alone, which is what makes the first six months survivable.
Decays on purpose as revenue arrives. Paying for presence forever buys a grid of machines that only prove they exist.
The demand side is subsidised directly, rather than by hoping supply attracts it.
115 MON per H100. A hundred of them is 11,500 MON earning continuously with nobody renting anything.
Capacity bought forward. It is how a provider scales without waiting for utilisation that does not exist yet.
Keyed to the fingerprint, not the token. You cannot mint a clean sheet by starting over.
The NFT system
One is welded to the silicon and can never be sold. The other is a tradeable claim on what that silicon earns. The split is the whole design: identity that cannot be traded away, and revenue that can.
A soul exists only after an attestor quorum watched that exact device pass a timed proof of compute. It carries the hardware reputation and the stake behind its promises, and it dies if the machine stops answering for 72 hours.
When a machine leaves, the deposit unlocks and the token stops being soulbound. It keeps every trait it rolled — same eyes, same mouth, same finish — rendered as a ghost of itself, and it gains one thing a live soul cannot have.
That is why the eight-cap does not brick a machine. Supply is exactly eight, hardware is never permanently excluded, and coming back requires owning a relic — which is precisely what makes a relic worth something to somebody other than the person who minted it.
A share of every job that machine settles, for as long as it is held, paid in the settlement token at settlement. Not a share of the protocol — a share of that machine, which you can inspect: its class, its capability, its uptime, its completed-job count.
Capital now against revenue later, capped at 25% so they always keep the majority, plus a 2.5% royalty on every resale forever. Eight licences maximum per soul, which is what bounds the settlement loop and keeps a payout affordable to compute.
Pay structure
| Who | Earns | In | For | Rate |
|---|---|---|---|---|
| Provider | Job payment | listing token | Work delivered and confirmed | 71%+ |
| Provider | Settled emission | $WRAITH | Per WCU-hour of paid work | 5.00 |
| Provider | Availability emission | $WRAITH | Per WCU-hour online and listed | 1.00 |
| Provider | Deposit yield | MON | Periods it completed work in | 70% of swept |
| Provider | Licence royalty | listing token | Every resale of a licence on its machine | 2.5% |
| Licence holder | Revenue share | listing token | Every job that machine settles | up to 25% |
| Licence holder | Referral | listing token | Sourcing a job for that machine | 1.5% |
| Renter | Rebate | $WRAITH | Spending on the grid | governance-set |
| Renter | Breach payout | listing token | A machine failing its residency proof | refund + bond |
| Juror | nothing | — | Ruling on disputes | unpaid today |
| Attestor | nothing | — | Verifying hardware | unpaid today |
| Treasury | Protocol fee | listing token | Every settled job | 2.5% |
| Treasury | Yield share | MON | Deposits staked in shMON | 30% of swept |
Jurors and attestors earn nothing, and that is on purpose. Both roles are run by WRAITHGRID rather than farmed out to whoever will take a fee. A verification service that is up, and a panel that actually rules, matter more right now than a fee schedule for roles nobody has volunteered for. It is an operating cost, not a missing incentive.
Attestors decide what hardware is real. Jurors decide who was right when a job goes wrong. Both are run by WRAITHGRID, deliberately, and that is a trade rather than an oversight.
A verification service that is actually up, and a panel that actually rules — on a consistent, current model rather than whatever an anonymous operator felt like running. A dispute system nobody staffs is worse than one staffed by a known party, because escrow sitting unresolved is a real loss to a real renter.
Threshold signatures across attestors we all run give redundancy, not independence. Three jurors on one operator is one opinion counted three times, whatever the median says. So the honest claim is not "trustless" — it is that the parts holding your money do not require trusting us, and the two parts that judge things do.
Both are governance-seatable and revocable, so this is a starting configuration rather than a permanent one. Removing an attestor instantly invalidates every unredeemed signature it produced — that is the revocation path, and it exists precisely because the role is trusted.
Supply
A fixed-supply memecoin already live on Monad — no mint function, no owner, no fee on transfer. Of the founding position, 100M is committed to seeding the grid.
Seeded supply is never trapped. Governance can withdraw anything not already owed to somebody, at any time, with no timelock — but nothing that has already been earned. That is the one restriction that has to exist for the promise to mean anything to a provider.
