Liquidation bots on Sui can spend a few thousandths of a dollar on a transaction, or hundreds of dollars competing for the right to execute it. The code has not become thousands of times more expensive. The opportunity has.
That difference is worth examining because much of the value in a conventional DeFi liquidation is assigned before the liquidator does any work. A protocol announces a collateral discount. Whoever gets the transaction through first collects it. When the discount is large and the execution cost is small, bots have a reason to spend the difference on transaction priority.
What would happen if liquidators competed on the price of collateral ahead of time, and anyone could execute the resulting trade when a loan crossed its liquidation threshold?
A second-price auction provides an interesting answer. Sui’s low transaction costs and Seal’s on-chain decryption support make the idea more practical than it would have been a few years ago.
1. Where the liquidation discount goes
Take a borrower — call her Alice — with $12,000 in collateral, $10,000 in debt, and an 80% liquidation threshold. Her health factor is 0.96, so her position can be liquidated.
Now a liquidator shows up — call him Bob. He repays $5,000 of Alice’s debt and receives $5,250 of her collateral. The debt drops to $5,000, the remaining collateral is worth $6,750, and the health factor rises to 1.08. The lender is less exposed, and Bob walks away with $250 in gross compensation.
Alice’s equity fell by exactly that $250. That is the price of the liquidation bonus, before any other fees.
The interesting question is whether the $250 was necessary. Liquidators like Bob differ in their routes to USDC liquidity, their inventory, their exchange access, their execution costs. One might need a $200 discount to bother; another would happily take $60. In a protocol with a fixed 5% liquidation bonus, every one of them faces the same terms. The competition decides who captures the $250, never how much Alice pays for the service.
That points at something the fixed-bonus design quietly bundles together. A liquidation is really two problems: repaying the debt, and finding a price for the collateral that changes hands. They are usually settled in one operation, but nothing says they have to be priced by the same mechanism.
2. How current protocols sell collateral
The simplest approach is a fixed incentive. Aave V3 allows a third party to repay a liquidatable borrower’s debt and seize collateral with the configured bonus. The operation is atomic and can be financed with a flash loan. Depending on the health factor and position size, V3 allows repayment of up to 50% or 100% of the debt. Morpho Blue follows the same immediate-repayment principle, with an incentive derived from the market’s liquidation loan-to-value threshold. A Morpho market at 86% LLTV, for example, gives an incentive factor of about 1.044.
These mechanisms are fast once somebody submits a successful transaction. The price is already fixed. A liquidator accepts it or leaves the opportunity to someone else.
Euler takes a different route. Its liquidation discount increases as a borrower’s health deteriorates. Aave V4 also makes liquidation terms depend on the position’s health, with a target health factor governing the amount repaid. The incentive responds to risk, although participants still accept terms calculated by the protocol rather than bidding against each other for a specific lot.
Sky’s Liquidations 2.0 uses a Dutch auction. When a vault is liquidated, the system starts an auction above the reference price and lowers the asking price according to a configured curve. Buyers call Clipper.take() when they find the price acceptable. A purchase settles immediately and can cover only part of the lot. The full auction may span many blocks.
There is a useful on-chain example. On April 7, 2025, three recorded take() calls for ETH-A auction #901 occurred in Ethereum blocks 22,215,754, 22,215,755 and 22,215,757:
| Purchase | Gas used | Transaction fee |
|---|---|---|
| First recorded take | 124,042 | 0.007511 ETH |
| Second take | 141,178 | 0.004828 ETH |
| Final take | 167,458 | 0.011454 ETH |
Those purchases used 432,678 gas in total and were separated by three blocks, or 36 seconds, from the first recorded take to the last. Their fees sum to 0.023793 ETH. This is the cost of the three listed purchases, not the cost or complete duration of the auction: the opening transaction and any unlisted activity are excluded.
Compound III separates the stages differently. Its reserves absorb the bad position through absorb(). The protocol then sells the collateral through buyCollateral() at an oracle-based discount. The borrower’s debt is dealt with before a buyer takes the collateral, while the reserve bears the intermediate exposure.
These examples lay out the design choices actually in use. Aave and Morpho settle at a known incentive. Euler and Aave V4 vary that incentive with risk. Sky discovers the purchase price over time. Compound funds the immediate debt absorption from reserves. What none of them does is the thing we’re after: keep the atomic settlement these mechanisms already have, but let the collateral buyers compete on price instead of handing every one of them the same fixed discount.
3. What the second price changes
In a first-price sealed-bid auction, the highest bidder wins and pays their own bid. A bidder who thinks a lot is worth $10,350 has to decide how far below that valuation to shade his bid. Bid low and he keeps more profit if he wins, but he’s likelier to lose outright. So he spends his effort guessing what everyone else will do — not telling the truth about what the collateral is worth to him.
A Vickrey auction removes that guessing game: the winner pays the second-highest valid bid. Suppose three buyers value a liquidated lot at $10,350, $10,250 and $10,100. If they submit those values, the first wins and pays $10,250. The winner keeps $100 of surplus against his own valuation, and with $10,000 of debt to repay, the sale leaves $250 for the borrower before separately funded execution costs.
That pricing rule is the whole point. Since the winner pays the highest competing bid, nudging your own bid up or down never changes what you pay when you win. The only thing your bid controls is whether you win at all.
Stated precisely: let a bidder value the lot at , and let be the highest competing bid. Winning is desirable when , and losing is desirable when . Since the winner pays , changing one’s own bid does not improve the payment on a winning outcome. Bidding is therefore a weakly dominant strategy under the independent-private-value assumptions of Vickrey’s original result.
Collateral liquidation is a more complicated setting than the textbook model, but the pricing rule is still useful. Each participant can incorporate their own inventory, swap routes, financing and expected proceeds into a bid. The auction selects the highest eligible offer without requiring the buyer to win a race to submit a liquidation transaction.
A Dutch auction also lets buyers decide what the collateral is worth to them, but the decision is made while the asking price falls. The bidder must choose when to take it. A second-price auction can instead calculate the result from bids already available at the time of liquidation. There is no descending-price period in the settlement path.
Academic work gives us a reason to care about the cost of this competition. In Liquidation Mechanisms and Price Impacts in DeFi, Phoebe Tian and Yu Zhu find that lower participation costs can make auction-based liquidations more competitive and reduce their price impact. The value of an auction depends partly on the cost of joining it.
That last point is particularly relevant on Sui.
4. What liquidations cost on Sui
We examined an October 2026 mainnet dataset containing 3,339 liquidation events across 2,803 successful transactions in six Sui protocols. Gas figures come from transaction effects: computation cost plus storage cost minus storage rebate, converted to USD using historical daily SUI prices. Events were deduplicated by transaction digest for gas statistics.
| Protocol | Successful transactions | Median net gas, USD | P95 net gas, USD |
|---|---|---|---|
| Suilend | 722 | $0.1043 | $5.0117 |
| NAVI | 556 | $0.0026 | $0.1096 |
| Scallop | 581 | $0.0025 | $0.0627 |
| Bucket V2 | 408 | $0.0164 | $0.2359 |
| DeepBook Margin | 287 | $0.0067 | $0.3199 |
| Current Finance | 249 | $0.0069 | $0.0138 |
These are observed costs; they are not a ranking of smart-contract efficiency.¹
Taken together, the transactions show what the network can execute cheaply. A Bucket V2 transaction on November 4, 2025 repaid approximately 446,444 USDB. Its 33 PTB commands included oracle operations, flash minting, collateral liquidation and DEX swaps. Net gas was 0.06177318 SUI, approximately $0.12 at that day’s price. Call that the floor: the real cost of doing the work.
Now watch what happens when execution priority is the prize instead.
On October 29, 2026, a NAVI liquidation repaid about $52,658 in debt and spent 458.5317 SUI on gas, approximately $398.69, at a gas price of 33,226,859 MIST per computation unit. That $398.69 bought transaction priority, not computation.²
On October 8, 2026, a Suilend liquidation repaid about $1,019. The event-derived gross liquidator bonus was approximately $30.57, and gas alone ate $25.70 of it, about 84%.³
Sui’s local fee markets explain the contrast. When bots compete to mutate the same shared object, they can bid up gas prices to win transaction priority. A cheap operation turns expensive not because the work got harder, but because several actors are chasing the same reward.
¹ The samples cover different time ranges. Many NAVI and Scallop events are micro-liquidations, and a single PTB may bundle several liquidations, oracle updates, or swaps. The dataset also excludes failed attempts and transactions never included on-chain.
² An ordinary computation of this shape costs cents (see the Bucket transaction above). The $399 is almost entirely a priority bid.
³ The 84% isn’t the liquidator’s net loss; swap costs and other cash flows still move in his favor. But it shows how much of a small bonus the gas race alone can consume.
A second-price liquidation changes what they compete for. Bidders compete through standing offers for the collateral. An executor receives a separate fee for submitting the liquidation transaction. The large discount on collateral no longer goes to whoever wins transaction priority.
There can still be competition for the execution fee, but the amount exposed to that race is much smaller and can be set independently of the collateral price. Cheap bid submission and cheap settlement make this separation economically attractive on Sui.
5. Standing bids with Seal
That separation only works if a bid can sit on-chain without anyone seeing it until it’s time to settle. Publishing ordinary Move fields would reveal every offer. A commit-reveal design keeps bids hidden at first but makes participants come back later to reveal them, an awkward dependency for liquidations, which fire exactly when nobody is watching.
Seal offers another approach. Its key servers release derived decryption material according to access rules defined in Move. For small encrypted values, seal::bf_hmac_encryption supports verification of derived keys and decryption inside a Move transaction. The project’s encrypted voting example demonstrates on-chain tallying after the required keys become available.
In the proposed liquidation market, buyers would submit encrypted offers for defined collateral lots or asset pairs. Every bid would be backed by purchasing funds held in escrow, so the winner does not have to be online when the opportunity appears. A standardized escrow cap per lot could support different encrypted prices without disclosing each exact bid through its deposit size.
An offer could specify the quantity it covers, its expiration and a price expressed relative to a reference oracle. That last choice matters for standing bids: a price quoted once in absolute USDC terms may become stale as SUI moves. At settlement, eligible offers can be normalized to the same lot and debt currency.
The executor gathers the required derived keys from Seal’s key servers before constructing the final transaction. In the PTB, the Move code verifies the keys against trusted server public keys, decrypts the eligible offers, identifies the two highest valid prices and performs the trade. The price calculation is verifiable on-chain. The key-serving step remains off-chain and is governed by the configured threshold policy.
One caveat: this is a design proposal, not a deployed implementation or a measured gas benchmark for Seal-based auctions. The capability it leans on is established. Sui provides on-chain decryption and atomic transaction composition, so a liquidation contract can consume decrypted bids directly during settlement.
6. A liquidation in one PTB
Put the pieces together and return to Alice. Take a simplified version of her position:
| Item | USD value |
|---|---|
| Collateral offered for liquidation | $10,500 |
| Debt to repay | $10,000 |
| Highest funded bid | $10,350 |
| Second-highest funded bid | $10,250 |
| Third bid | $10,100 |
Assume Alice’s position is eligible for liquidation, the three bids cover the same lot, and all settlement assets are available. The highest bidder wins; the amount payable for the collateral is $10,250.
This is where Bob comes back, now as the executor rather than the liquidator racing for priority. He (or anyone else willing to do the job) obtains the Seal key material and submits a PTB that validates Alice’s position and the eligible bids, computes the winning bid and the second price, repays $10,000 of her debt, and transfers the collateral to the winner. The payment comes from the winner’s escrow. The remaining $250 of sale proceeds goes back to Alice.
If the lending contract or auction interface needs temporary liquidity before the escrowed payment is released, the executor can take a flash loan at the start of the PTB and repay it from the sale proceeds before the transaction ends. DeepBook’s flash-loan interface shows how this form of atomic financing already works on Sui. Where escrowed funds can pay the loan directly, no flash loan is needed.
Bob’s compensation is separate from the auction entirely. One workable rule: reserve a small fraction of each loan when it’s opened. The reserve pays out in proportion to the debt actually repaid in a partial liquidation, and anything unused returns to the borrower when the loan closes. For a 0.1% fee on $10,000 of repaid debt, Bob earns $10 from that reserve, not a surcharge skimmed off the winning bid.
That’s what keeps the second price honest: the winning bidder pays $10,250 whether he executes the liquidation himself or Bob does it for him. Unused escrow is released after settlement.
Alice’s economics are the punchline. A fixed 5% bonus on a $10,000 repayment would hand over $10,500 in collateral and leave her nothing from this lot. Under the auction, the sale returns $250 after settling her debt. Net the separately reserved $10 execution fee and Alice walks away $240 better off than the fixed-bonus world would have left her, with all other costs excluded.
A successful settlement requires one PTB. Preparing and funding bids happens beforehand; obtaining Seal keys happens before submission. No winner needs to confirm the purchase, and there is no sequence of Dutch-auction price reductions after liquidation becomes available.
7. A different market for liquidations
None of the six Sui protocols we measured runs an auction at all. Every one settles at a fixed or risk-adjusted bonus inside a single atomic PTB.⁴ So this isn’t a tweak to an existing Sui auction; it’s a different market structure for a chain that doesn’t yet have one.
The core difference from Aave or Suilend is who prices the collateral. In the familiar fixed-bonus design, the protocol picks the reward and bots compete for it. In the proposed market, buyers compete through funded bids and the contract applies a second-price rule. The borrower keeps whatever remains after the debt is settled.
Compared with Sky, the bid collection takes place before liquidation. A liquidatable position can therefore be settled without waiting for a Dutch price curve to descend. Compared with Compound III, the purchase is financed by committed bidders rather than relying on protocol reserves to hold collateral between absorption and sale.
The executor still has a reason to act quickly: a timely liquidation earns the reserved fee. The reward can be kept modest because it no longer includes the entire collateral discount. That limits the value available to spend on transaction priority while retaining permissionless execution by multiple independent bots.
The case for Sui rests on actual transaction economics. Ordinary liquidation PTBs can cost cents or less; priority competition can cost much more. Seal adds a means of handling confidential bids, and PTBs allow the final debt repayment and collateral transfer to settle together.
The proposal comes down to two moves: let liquidators compete on the collateral price before a liquidation happens, and pay separately for the transaction that closes the debt. A second-price rule makes truthful bidding the dominant strategy, and because settlement on Sui costs cents, splitting the two prices apart doesn’t eat the value the split is meant to recover. Under a fixed bonus, Alice’s $250 goes to whichever bot wins the gas auction. Here it goes to the bidder who valued her collateral the most.
⁴ Across the October 2026 dataset (Suilend, NAVI, Scallop, Bucket V2, DeepBook Margin, and Current Finance), every liquidation settles debt and seizes collateral in one atomic transaction at a fixed or risk-adjusted bonus. The one auction-like module on-chain, Current Finance’s time-decaying auto-deleverage path, has never fired on mainnet.
Sources and transaction data
- William Vickrey, Counterspeculation, Auctions, and Competitive Sealed Tenders (1961).
- Phoebe Tian and Yu Zhu, Liquidation Mechanisms and Price Impacts in DeFi (Bank of Canada, 2025).
- Official documentation: Aave, Morpho, Euler, Sky, Compound III, Seal.
- Sui mainnet research dataset, collected October 31, 2026: 3,339 liquidation events, 2,803 successful transactions across six protocols. Transaction digests for the cited Sui examples are linked in Section 4. USD amounts use historical daily prices; failed transactions and nonincluded attempts are not in the sample.