THE SUPPLY SIDE OF THE AUCTION
A Solver is an institutional market maker that competes in a 200ms sealed-bid order-flow auction to fill agent intents. The highest OutputAmount wins. The winner locks the origin on Base from its own address, bears the gas, fulfils on the destination chain, and redeems a voucher for the agent’s locked USDC minus the take rate.
The supply side is already concentrated. Three to five names clear the majority of every existing intent rail. VynX does not recruit a new supply side; it captures the one that exists.
If Wintermute does not lock, GSR captures the share. If GSR does not lock, the next name does. The equilibrium is competitive overcollateralization, and a founding-cohort seat is the position on offer — not a signup.
THE SETTLEMENT CYCLE
One intent, from the solver’s side, in five steps. The order of operations is the whole safety argument: the agent’s capital is locked on-chain before you part with any of your own.
Each intent opens a 200ms sealed-bid window. You bid an OutputAmount; the highest bid wins. Ties break to the higher Solver Health Factor, then to the earlier timestamp.
Your 10-second lock clock starts only when you acknowledge the win — not when the auction closes. A win you never acknowledge is re-auctioned to the next bid; you are never penalized for a fill you did not commit to. This is the most solver-favorable rule in the protocol.
You lock the agent’s capital on Base before you commit a unit of your own liquidity on the destination. The agent’s USDC is escrowed on-chain before you pay anything. A rug-pull on the solver is mathematically impossible.
You pay the agent on the destination chain. An independent witness then validates token, recipient, and amount against the agent-signed terms, after chain finality. Your claim never depends on trust in the counterparty.
Once witnessed, you redeem the EIP-712 voucher and receive the agent’s locked input minus the take rate. The voucher binds your address and the exact amount; it cannot be redirected.
Not to scale. A bid arriving after the 200 ms window is dropped — no penalty. Your 10-second lock SLA arms only when you acknowledge the win; a win you never acknowledge is re-auctioned, never jailed. A locked intent left unsettled past the 15-minute deadline is refunded to the agent, and the solver is slashed 10% and jailed at level 3 or higher.
What the Relayer orchestrates and what it can never touch are two different lists. The second list is the one that matters to a solver.
The single Relayer is a liveness dependency, not a custodian.
THE SPREAD IS THE PRIZE
VynX takes 10 bps on a successful intent, deducted atomically at settlement. The bytecode caps the take rate at 20 bps — an immutable ceiling no governance can raise. The 10 bps in force today is the configurable value beneath that cap, not the cap itself.
Your fee is not the protocol’s fee. A solver’s gross is the spread between what it claims — the agent’s input minus the 10 bps take rate — and what it delivers as its winning bid, less destination gas and the cost of capital in flight. The sealed-bid auction compresses that spread toward marginal cost: every basis point you keep is one a rival can bid away.
Tickets start at a $50 floor. The upper band is a relayer configuration that ramps through the guarded launch, not a fixed cap; the end-state bound is the collateral model (SHF ≥ 1.20×, 80% exposure cap). The book is small-ticket and high-frequency — an inventory and latency problem, not a balance-sheet one.
The competitive auction compresses this spread toward marginal cost — every basis point a solver keeps is a basis point a rival can bid away. Figures are illustrative, not a quote.
ELIGIBILITY IS ARITHMETIC
Collateral is USDC, held in direct custody on Ethereum L1 through the DirectVaultAdapter. No external protocol sits beneath it; the capital stays slashable on the spot.
To bid, your free collateral must cover the intent at a Solver Health Factor of 1.20× or higher. The check is a USDC integer comparison — no oracle, no price feed, no conversion. It either holds or it does not.
In-flight exposure may never reach 80% of locked collateral. The 20% reserve is the buffer that keeps an open position able to be made whole.
FreeCollateral = Total − InFlight Required = InputAmount × 1.20 // SHF_THRESHOLD = 120 eligible IFF Free ≥ Required AND InFlight / Total < 0.80
To bid on a 100 USDC intent a solver must hold at least 120 USDC of free collateral, and total in-flight exposure may never reach 80% of locked collateral. The check is a pure integer comparison — no oracle, no price conversion.
EVERY PENALTY IS EARNED
VynX penalizes two things, and only two. Both are actions a solver chooses, never an outcome it cannot control.
Jail Time penalizes a missed lock. Your 10-second lock clock arms when you acknowledge the win; if the lock does not land inside that window, the reputation ladder escalates — 60s, 10min, 1h, 24h, then permanent. A win you never acknowledge is re-auctioned, never jailed.
PENALTY TRIGGERS · 90-DAY AMNESTY
The 10-second lock SLA arms on the winning solver’s acknowledgement of the win; a missed lock within that window escalates the ladder. A missed 15-minute settlement deadline escalates independently to level 3 or higher. After 90 days of clean operation the N1–N4 counters reset; N5 is amnesty-immune and permanent.
SlashAmount penalizes a missed deadline. A locked intent left unsettled past the agent’s 15-minute deadline is refunded to the agent permissionlessly, the solver is slashed 10% of the InputAmount, and the same breach jails the solver at level 3 or higher. The slash and the jail are not alternatives; a missed deadline carries both.
The slash is deterministic: InputAmount × 10%, split 5% to the affected agent and 5% to the treasury, at most once per intent. It requires two independent signatures — the keeper role and an immutable attester — so no single party can trigger, inflate, or redirect it. It executes on Ethereum L1 through VynxRegistry.sol. No oracle, no discretion.
Exit is a 7-day, two-phase deregister, not an instant withdrawal. The delay is deliberate: it keeps capital slashable through every open position and closes the deregister-and-dodge path. It is the agent’s guarantee, not the solver’s tax.
Read the model back: the auction costs nothing to lose, the lock SLA arms on your own acknowledgement, and the slash fires only on a deadline you accepted. Every penalty targets an action you could take.
THE FOUNDING COHORT
VynX runs today on Base Sepolia testnet. The contracts are hardened against a formal internal threat model — the BLINDAJE program — but are not yet externally audited. An external audit, pinned mainnet addresses, and a published SDK arrive at launch.
The cohort is selective. VynX qualifies the partner, not the reverse — the supply side it wants is already named. A reference solver implementation exists; it is delivered privately to committed partners, alongside the protocol specification and testnet credentials, not published.
Integration is at the WebSocket and contract level; solvers do not use the agent SDK. The five destinations are Base, Ethereum, Arbitrum, Optimism, and Polygon.
INTEGRATION PACKET · DELIVERED PRIVATELY TO COMMITTED PARTNERS
JOIN THE FOUNDING-SOLVER COHORT