Arbitrum Open House Singapore · Buildathon
Exit Market
The safest way out of Arbitrum. Now instant.
Scroll, or use the arrow keys
The problem
6.4 days
Leaving Arbitrum the safe way takes a week. Once a withdrawal has started, nothing can speed it up.
Challenge period: 45,818 L1 blocks on Arbitrum One and mainnet Orbit chains.
Arbitrum One → Ethereum · snapshot 2026-09-29
The money that waits
Tokens only, from chain data, reproducible in research/stranded. Native ETH bypasses the token gateway unless withdrawn as WETH.
Why it is still unsolved
Nobody rescues a withdrawal in flight
Fast bridges
A route you pick before withdrawing, with its own relayers, oracles or attesters. Across lists none of Xai, ApeChain, RARI, Sanko or EDU Chain.
Native fast withdrawals
A chain-wide validator committee. Per Arbitrum's docs, such a chain "would technically no longer be a Rollup".
Exit Market
Per withdrawal, after it started, on the chain as deployed. No committee, oracle or relayer decides validity.
The insight
The hook was already there
L1ArbitrumExtendedGateway.Every Arbitrum token gateway can hand a pending withdrawal to a new owner. Its source says: "It is assumed the _exitNum is validated off-chain." So nobody could safely buy one. Exit Market is that missing validation, on-chain.
How it works
One signature instead of a week
- 01
Withdraw
A standard bridge withdrawal on the child chain
- 02
Prove
Leaf and Merkle path against a still-pending root
- 03
Sell
The vault pays USDG now and owns the exit
- 04
Collect
A permissionless keeper executes; LPs earn the discount
Sellable about 15 minutes after withdrawing, when the next rollup node posts, not 6.4 days later.
Verified on-chain, in the seller's transaction
Only Arbitrum's own commitments
Market owns the exit
gateway.getExternalCall(exitNum, initialDestination)Leaf is this withdrawal
Outbox item rebuilt byte for byte, minimal Merkle pathRoot is real while pending
Legacy: node.confirmData · BOLD: assertion chain · both: no rival at any pending levelNot claimed yet
!Outbox.isSpent(index)No oracle, no committee, and the market has no owner. The buyer's risk is a pending rollup node being rejected, which the vault prices and writes off; the moment a validator disputes it, it stops trading.
Gasless exits
No ETH on the other side? Sign once.
- 01 · Withdraw on the child chain straight to the ExitIntentRouter.
- 02 · Sign one EIP-712 sell order. No gas, no transaction.
- 03 · Any relayer settles it into the vault and earns a small fee.
Live · exit #7
0 ETH
held by the seller on Arbitrum Sepolia, who still received the USDG.
Live on testnet · Xai Testnet → Arbitrum Sepolia
Every step is a real transaction
A 2.5 USDG withdrawal leaves Xai Testnet
0x15e38d3a…5488 ↗Sold while still pending, in one signature
0xa6eb8d37…bf55 ↗Gasless: settled for a wallet holding 0 ETH
0xdcabb327…2f57 ↗Listed at the seller's price, bought by another wallet
0x6ff67081…7031 ↗Executed through the Outbox, face value collected
0x69afb2b2…8181 ↗Use of Arbitrum technology
Built on the protocol, not beside it
Token bridge
transferExitAndCall on the real parent gateways
Outbox
Leaves rebuilt byte for byte; NodeInterface proofs
Orbit L3
Live end to end on Xai Testnet
BOLD
A real pending Arbitrum One exit proven over its real pending chain (mainnet fork, ~1.3M gas)
Stylus
The proof core in Rust, live, with an honest gas benchmark
Rollup nodes
Pending legacy nodes already commit to their send roots
Security and quality
Every high finding fixed, with a regression test
Round four found a critical balance-delta bug in the first router; round five found the same pattern in the market and an owner key that could allow a hostile gateway; round six found the legacy verifier still trusted a node after a validator disputed it, and a buyer contract could be pulled into a sale it never agreed to. Each was reproduced as an exploit test, fixed and redeployed; the live market now has no owner. AI-assisted internal reviews, not a third-party audit.
Competition
Rescue an exit already in flight
| Option | Works after withdrawing? | Trust for validity | Orbit L3s |
|---|---|---|---|
| Exit Market | Yes, per withdrawal | Rollup commitments only | Any, once deployed for its gateways |
| Native fast withdrawals | Yes, if the chain opts in | Validator committee | Opt-in per chain |
| Across | No | Relayers + UMA oracle | None of the five |
| Circle CCTP | No | Circle attestation | USDC only, no L3s |
| Stargate / Relay | No | Pools / solvers | ApeChain only |
Sources checked 2026-09-29. On price, CCTP and Stargate are cheaper on the routes they serve; we do not compete there.
Business model · first 30 days on mainnet
Idle capital becomes yield
Revenue
A 0.25% market fee to the protocol. The vault discount, about 0.27% for a full window, goes to LPs: roughly 15.7% APR gross at full utilisation.
KPIs
20+ exits from 10+ sellers, $100K+ face value. $50K+ vault TVL at 40%+ utilisation with zero write-offs. 5%+ of eligible in-window value bought.
All measurable on-chain from ExitVerified, sale events and vault totalAssets.
What comes next
From Offchain Labs research to the deployed bridge
Moosavi, Salehi, Goldman & Clark (AFT 2023) showed tradeable exits on a modified Nitro. Exit Market runs on the bridge that is already live. Next: Arbitrum One → Ethereum on mainnet with the BOLD verifier, one vault per asset, and an instant-exit button in Orbit chains' own bridge UIs.