On 12 March 2020, while ether lost a large part of its value in a day, the Ethereum network was so congested that most of Maker’s liquidation bots could not get their bids in. For two to three hours, according to the community’s own response thread on the Maker forum, a single keeper kept winning collateral auctions with bids of zero dollars, with no one to outbid it. Borrowers lost their ether for nothing, in what is still the best-known failure of DeFi liquidations. The system, which had a surplus of about 500,000 dollars before the crash, ended the day with a hole of about 4 million.
Almost six years later, on 5 February 2026, Aave closed a week in which it had processed 429 million dollars of liquidations across roughly 12,500 transactions, a record for the protocol in dollar terms, and its own blog says it did so with no downtime or system failure. Same job, same kind of chain, very different outcome. What changed sits in the plumbing of liquidations: who runs them, how the price is set, how the borrower is charged and who pays when the collateral is not enough.
This lesson follows on from the previous one, how the health factor and the liquidation price are calculated. In one sentence: your position becomes liquidatable when the health factor, the value of your collateral weighted by its liquidation threshold divided by your debt, falls below 1. What happens at that moment is the subject here. If the smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool model, aTokens and eMode are new to you, start with our guide to crypto lending and borrowing.
How DeFi liquidations work: a sale nobody has to approve
A liquidation in wallet.">decentralised finance is not a decision. No risk officer calls you, no court is involved and there is no grace period. The lending protocol publishes a function anyone may call. When an account’s health factor is under 1, a third party repays part or all of that account’s debt with its own tokens and, in exchange, receives the borrower’s collateral worth the repaid amount plus a discount. The protocol only checks the arithmetic.
That design solves a problem a bank solves with staff. A protocol holding billions of dollars of loans cannot sell collateral by itself: it needs someone to bring the debt asset, take the collateral and carry the market risk of reselling it. The discount pays for that service. On Aave V3 it is called the liquidation bonus, and the Aave Help Center puts it simply: the liquidator is repaid in collateral at full value, and the bonus comes on top, out of the borrower’s side.
Speed is the second ingredient. Loans in DeFi are overcollateralised, so in normal markets a liquidation repays a debt that is still fully covered. The whole system depends on it happening while the collateral is worth more than the debt plus the bonus. Once prices fall faster than liquidators can act, the safety margin is consumed and the protocol starts to lose money. That race decides most of what follows.
Who liquidates: bots competing for a discount
The bonus, the race and the gas
Liquidators are almost always programs. They read every borrower’s position from the chain, recompute health factors as oracle prices update and send a transaction the moment an account crosses the line. On Aave V3 the entry point is a function called liquidationCall: the bot names the collateral it wants, the debt asset it repays, the borrower and the amount of debt to cover, and it can choose to receive the collateral as aTokens rather than the underlying asset. Nothing restricts who may call it.
Because the bonus is public and the same for everyone, profitable liquidations attract competition. Several bots usually spot the same account in the same block, and only the first transaction to be included gets the reward. On Ethereum that turns liquidations into a bidding contest for block space: bots raise their fees or send private bundles to block builders so their transaction lands first. Part of the bonus therefore leaves as fees to validators and builders, and the borrower pays the full discount either way.
Academic work on the first DeFi boom found the same imbalance. A study by Kaihua Qin and co-authors, An Empirical Study of DeFi Liquidations, looked at Aave, Compound, MakerDAO and dYdX, which it says together held over 85 percent of Ethereum’s lending market, and concluded that existing designs do pay liquidators enough but sell more discounted collateral than needed at the borrowers’ expense. That finding still frames the debate: a bonus large enough to attract bots in a crash is a cost to every borrower in calm markets.
Protocols pay liquidators in other ways too. Maker, now Sky, gives the keeper that starts an auction a flat fee and a percentage of the debt, both drawn from the system’s surplus. Compound III keeps a points ledger for each liquidator, counting successful calls and gas spent, and its documentation says compensation may come through governance in the future. Liquity V1 paid the caller a fixed gas compensation in its stablecoin plus a small share of the collateral.
Flash loans: liquidating with no capital
A liquidator needs the debt asset up front: to repay 5 million USDC of someone’s loan, it must hold 5 million USDC for the length of one transaction. Flash loans remove that requirement. A flash loan is a loan with no collateral that must be repaid inside the same transaction; if it is not, the whole transaction is cancelled as if it never happened. A later lesson of this course covers them in detail.
Inside a liquidation the sequence fits in one transaction: borrow the debt asset by flash loan, call the liquidation function, receive the collateral, sell it on a decentralised exchange, repay the flash loan and keep what is left of the bonus. Sky’s documentation describes the same idea for its own auctions: a bidder can route the collateral into other protocols during the purchase to produce the DAI it owes, so it needs only gas to take part. No public dataset we could verify gives the share of liquidations funded this way, so we do not quote one.
Four designs for the same forced sale
Every lending protocol has to answer the same questions: who triggers the sale, at what price, how much of the loan can be closed at once and what happens if no one buys. Four answers dominate today.
Aave V3: a fixed bonus and a 50 or 100 percent cut
On Aave V3 each asset carries its own liquidation bonus, set by governance. The bonus is applied to the debt repaid: with a 5 percent bonus, a liquidator who repays 1,000 dollars of debt receives 1,050 dollars of collateral. A share of that bonus, the liquidation protocol fee, goes to the protocol’s treasury instead of the liquidator; it is taken out of the bonus, not added to the borrower’s bill. In the Chaos Labs report on the October 2025 crash, about 10 million dollars of bonuses produced about 1 million of fees for the Aave DAO, roughly a tenth (our division).
How much can be closed depends on how sick the account is. The liquidation logic in Aave V3’s source code sets a default close factor of 50 percent: while the health factor is above 0.95 and both the collateral and the debt are worth at least 2,000 dollars, a liquidator may repay at most half of the debt in that asset. At 0.95 or below, or on smaller positions, the whole debt can go. A partial liquidation must also leave at least 1,000 dollars of both collateral and debt; otherwise the position is cleared in full, so that no dust remains that nobody would bother to liquidate.
These are the parameters we read from Aave V3’s Ethereum Core market contract on 29 October 2026. They change by governance vote, and other Aave markets and networks have their own values.
| Asset (Aave V3 Ethereum Core) | Max LTV | Liquidation threshold | Liquidation bonus | Status |
|---|---|---|---|---|
| WETH | 81% | 84% | 5% | collateral and borrowable |
| WBTC | 81% | 85% | 5% | collateral and borrowable |
| wstETH | 79% | 82% | 6% | collateral, not borrowable |
| USDC | 75% | 78% | 4.5% | collateral and borrowable |
| LINK | 0% | 71% | 7% | no new borrowing power |
| CRV | 0% | 41% | 8.3% | frozen |
Source: on-chain read of Aave V3’s PoolDataProvider on Ethereum, 29 October 2026. eMode categories carry a bonus of their own, set separately from the asset’s. The bonus climbs as the asset gets thinner and more volatile, because liquidators must be paid more to take a token they may struggle to resell. The CRV line is a leftover of the story told further down.
Aave V4 changes the price rule. Its developer documentation says bonuses follow a Dutch auction in which a lower health factor gives a higher bonus, so a barely unhealthy account pays less than one deep under water. The leftover rule survives: if less than 1,000 dollars of debt would remain in a reserve, the whole debt in that reserve must be liquidated.
Compound III: the protocol absorbs, then sells
Compound III splits the job in two. First, any address may call absorb on an account that is under water. The protocol takes over the debt and the collateral at once, repays the debt out of its own reserves of the base asset, and credits the borrower with the value of the collateral minus a penalty set per asset, the liquidation factor. Second, anyone may call buyCollateral to purchase the seized collateral from the protocol at a discount to the oracle price, paying in the base asset.
The discount comes from a formula in Compound’s documentation: a global store-front price factor multiplied by one minus the asset’s liquidation factor. A rule caps the process: once the protocol’s reserves reach a governance-set target, collateral sales stop. The design has one advantage for borrowers. Liquidators do not race to grab their collateral directly; the protocol absorbs first and sells later, and the reserve takes the price risk in between.
Sky: Dutch auctions that settle at once
Maker’s response to Black Thursday was to replace its old auctions, in which bidders locked DAI and waited for a better offer, with a system it calls Liquidations 2.0. On Sky, the protocol’s current name, a keeper calls a contract named Dog to bark at an unsafe vault. The vault’s whole debt and collateral are seized at once and an auction starts in the Clipper contract for that collateral type.
The auction starts above the oracle price and the price then falls along a curve chosen by governance until someone accepts it. The buyer calls take with the most it is willing to pay per unit and receives the collateral immediately: Sky’s documentation says the Dutch auctions settle instantly, so no capital sits locked waiting for a result. If too much time passes or the price drops too far, anyone can restart the auction. The borrower pays a penalty called chop, set per collateral type; the documentation’s example is a chop of 1.13, which means a 1,000 DAI debt can raise up to 1,130 DAI from the sale. Remaining collateral goes back to the vault owner.
Liquity: a stability pool buys the debt
Liquity takes the bots out of the pricing. Holders of its stablecoin deposit it in a Stability Pool. When a loan, called a trove, is liquidated, the pool burns stablecoins equal to the debt and its depositors receive the collateral, worth 105 percent of that debt in Liquity V2. Liquity’s documentation states the borrower’s side: a liquidated borrower usually pays a 5 percent penalty and can claim back the collateral that remains. In V2 a trove becomes liquidatable when its loan-to-value passes 90.91 percent for ETH and 83.33 percent for staking means depositing a proof of stake coin with a protocol that stakes it for you and issues a transferable token (such as Lido stETH or rETH) representing the deposit plus rewards, which you can use or sell while the coins stay staked.">wstETH and rETH.
If the pool is empty, two fallbacks apply. A liquidator can bring the stablecoin itself and receive 105 percent of the debt in collateral, or the debt and collateral are redistributed to the other borrowers of the same market, whose debts rise in proportion. The second option means borrowers, not a treasury, carry the shortfall.
Morpho and crvUSD in brief
Morpho markets pair one collateral with one loan asset and let a liquidator repay up to 100 percent of the debt in a single transaction. The bonus, called the liquidation incentive factor, is computed from the market’s liquidation loan-to-value with a cap of 15 percent; Morpho’s documentation gives about 5 percent for a market at 86 percent, and says the protocol takes no fee on it. Curve’s crvUSD goes the other way and sells collateral gradually in a band of prices before any hard liquidation; that soft liquidation is explained in our lesson on CDP stablecoins and is not repeated here.
| Aave V3 | Compound III | Sky (Liquidations 2.0) | Liquity V2 | |
|---|---|---|---|---|
| Who triggers | any address, liquidationCall | any address, absorb | a keeper, Dog.bark | anyone; the Stability Pool absorbs |
| Who gets the collateral | the liquidator | the protocol, then buyers via buyCollateral | the auction winner via take | Stability Pool depositors |
| Price to the buyer | fixed bonus per asset (4.5% to 8.3% on the table above) | discount from a store-front formula | falling auction price | collateral worth 105% of the debt |
| Share of debt per call | 50%, or 100% at HF 0.95 or below and on small positions | whole account | whole vault, partial only if a global debt limit is reached | whole trove |
| If collateral falls short | deficit recorded, covered by Umbrella or treasury | protocol reserves | historically MKR debt auctions | redistribution to other borrowers |
What happens if you are liquidated: the borrower’s bill
You do not lose the whole position by default. You lose the collateral that was sold to repay the debt, plus the bonus or penalty on that part, and your debt falls by the amount repaid. What remains stays yours, still deposited and still borrowing. On Aave V3 a first liquidation often closes only half of the debt.
A worked example with example prices
Take an Aave V3 borrower with 5 ETH deposited and 10,000 USDC borrowed, using the WETH parameters in the table (liquidation threshold 84 percent, bonus 5 percent). The prices below are examples chosen for round numbers, not market quotes. At an example price of 2,300 dollars per ETH, the collateral is worth 11,500 dollars and the health factor is 11,500 x 0.84 / 10,000 = 0.966.
That is under 1 but above 0.95, and both sides exceed 2,000 dollars, so the close factor is 50 percent. A liquidator repays 5,000 USDC and receives 5,000 x 1.05 = 5,250 dollars of ETH, about 2.28 ETH at the example price. The borrower now has 6,250 dollars of collateral and 5,000 dollars of debt; the health factor is back at 6,250 x 0.84 / 5,000 = 1.05. The direct cost of the event is 250 dollars, the bonus, part of which goes to Aave’s treasury as the protocol fee. Gas, slippage and interest are ignored here.
The position is alive but fragile. If ETH keeps falling, it can be liquidated again a few blocks later. Now suppose instead that the drop was sharper and the collateral is worth 11,300 dollars at the moment the bot acts: the health factor is 0.949, at or below 0.95, and the whole 10,000 dollars of debt can be repaid in one call. The liquidator takes 10,500 dollars of ETH and the borrower keeps 800 dollars of it.
In the worst case a price gap leaves the collateral at 9,000 dollars for the same 10,000 dollars of debt. A liquidator can take all 9,000 dollars of collateral, but the bonus math no longer works: to be paid in full, the collateral would need to be worth at least 10,500 dollars. Whatever debt is left unpaid becomes the protocol’s problem, the bad debt of the next section.
Taxes: a forced sale is still a sale
A liquidation sells your collateral, and in most systems a sale is a taxable event. In the United States the IRS’s virtual currency FAQ says gain or loss is recognised when you sell or exchange virtual currency for other property, reported on Form 8949; it does not address DeFi collateral directly, so the treatment of a forced sale on a lending protocol is likely a disposal at fair market value, to be confirmed with a tax adviser. In the United Kingdom HMRC’s Cryptoassets Manual has a section on DeFi lending and staking, and Capital Gains Tax applies to disposals in general. Keep the transaction hash of every liquidation: it gives the date, the amount of collateral sold and the debt repaid. Borrowing strategies and their tax treatment country by country are the subject of the next lesson.
Bad debt: when the collateral no longer covers the loan
How a deficit is born
Bad debt appears when an account’s debt is larger than everything a liquidator can take from it. On Aave V3 the code handles it explicitly: if a liquidation consumes all of a borrower’s collateral and debt remains, that remainder is written off the account and added to the reserve’s deficit, and the contract emits an event called DeficitCreated. Lenders in that reserve are then owed more than the protocol holds until someone fills the gap.
Three things produce it: prices that jump past the liquidation zone, too little market depth to resell the collateral, and oracle prices that lag the market. They tend to arrive together, often on an asset with thin liquidity. Good oracle design matters as much as the bonus; our advanced guide to DeFi oracles explains how feeds are built and manipulated.
CRV on Aave, November 2022
The clearest case on Aave involved CRV, the token of Curve Finance. In November 2022 a trader, whom several outlets linked to an earlier exploit of another protocol, deposited USDC on Aave V2, borrowed about 40 million CRV according to Decrypt and sold it, in what Gauntlet, then a risk manager for Aave, called an attempt to squeeze the token. CRV first fell, then rose sharply, and the position was liquidated on 22 November 2022. Not everyone accepts the squeeze reading. On the Aave governance thread opened that day, a forum user posting as digi7287 wrote on 28 November: “The CRV attack wasn’t an attempted short squeeze, it was an attack on the reported oracle price.” That remains one participant’s reading, not a finding of Aave’s risk managers.
Liquidators could not buy enough CRV quickly to repay the whole debt, and Aave was left with about 2.6 million CRV of bad debt, the count given later by the data firm Kaiko. In dollars, the opening post of the governance thread put it at around 1.5 million, and Blockanalitica’s analysis reported by Decrypt at 1.6 million; the gap mostly reflects the CRV price used. Gauntlet’s statement quoted by Decrypt: “The attempt to squeeze CRV on Aave has been unsuccessful and unprofitable.” Aave then froze several thin markets, and on its Ethereum Core V3 market CRV remained frozen on 29 October 2026.
CRV again: a near miss in 2023, liquidations in June 2024
CRV came back into the spotlight after the exploit of several Curve pools in July 2023, a story told in our lesson on DeFi composability. Curve’s founder, Michael Egorov, had borrowed large sums against his own CRV on several lenders, including Aave. As the token fell, he sold CRV over the counter to investors and repaid; Decrypt later wrote that he had to repay 9.2 million dollars in Aave loans at the time. We found no source reporting a liquidation of those loans on Aave or bad debt for Aave in 2023: it was a near miss, unwound by negotiation.
The liquidation came on 13 June 2024. CRV fell to an all-time low of 0.23 dollars, and according to blockchain analysts at Arkham, cited by Decrypt, Egorov’s CRV collateral, worth about 141 million dollars the night before and securing about 95.7 million dollars of stablecoin loans, was liquidated across five protocols. Two of his accounts on Llamalend, Curve’s own lending market, were left with more than 1 million dollars of bad debt, still according to Arkham via Decrypt, which reported that Egorov cleared it after receiving 6 million dollars in USDT from an investor, Christian Seale. Other estimates of Llamalend’s losses circulated that week; we could not reconcile them with a primary source, so we quote only this one.
Who pays the hole
Each protocol keeps a different order of payers. On Aave V3 the first line is Umbrella, introduced in June 2025: users stake aTokens or GHO for a given asset and network, and according to Aave’s Umbrella documentation, slashing is triggered automatically when a deficit in that pool exceeds a set offset; the documentation’s example is that the DAO covers the first 100,000 USDT of bad debt before stakers are touched. Stakers face a 20-day cooldown before withdrawing and remain slashable meanwhile. Aave’s blog put Umbrella above 250 million dollars of staked assets in February 2026, with about 200 million in the treasury as a second reserve.
On Aave the treasury stands behind Umbrella. Elsewhere, Compound III draws on protocol reserves, Liquity spreads the loss over other borrowers, Morpho vaults pass it to depositors through a lower share price, and Maker historically printed and sold new MKR. That last method was first used after Black Thursday.
Cascading liquidations: when forced sales feed the fall
A cascade is a loop. Prices drop, positions cross their thresholds, liquidators sell the seized collateral on the market, the selling pushes prices lower and more positions cross their thresholds. When the chain is congested or oracles lag, the loop runs faster than the system can clear it. Two episodes six years apart show both outcomes.
Black Thursday, 12 March 2020
Maker’s collateral auctions in 2020 required bidders to post DAI and wait for the auction to close. On 12 March, as ether’s price collapsed, gas prices soared and many keepers failed to get their bids included. The Maker community’s response thread records that zero-dollar bids kept winning with no competition for about two to three hours, until other keepers came back, that the system’s surplus went from about +500,000 dollars to about -4 million, and that MKR was down 44 percent in 24 hours. Whether congestion alone explains the zero bids, or whether some bidders kept others out on purpose, is still debated.
To refill the system, Maker minted and auctioned MKR. The auctions ran from 19 to 28 March 2020 and sold 20,980 MKR for 5.3 million DAI, according to The Block, which reported they were intended to cover about 4.5 million dollars of undercollateralised debt; the venture firm Paradigm won 68 percent of the 106 auctions. The DAO later replaced the old auctions with the Dutch auctions described above. The wider history of that period is in our history of DeFi.
10 and 11 October 2025: derivatives venues and Aave are different products
One of the largest liquidation days on record began on Friday 10 October 2025, after the US president announced 100 percent tariffs on Chinese imports. Within 24 hours, more than 19 billion dollars of leveraged positions were liquidated, 19.37 billion across 1.6 million traders according to CCN, which cites CoinGlass for 7 billion in the first hour alone. Those are mostly perpetual futures: margin positions closed by exchanges and on-chain perpetual platforms when a trader’s margin runs out.
Lending liquidations are something else: a loan repaid with a sale of collateral, on a protocol like Aave. Chaos Labs, Aave’s risk manager, reported in its review of 10 and 11 October about 180 million dollars of on-chain liquidation volume on Aave, while Aave’s blog counts more than 250 million liquidated on 10 October. The two families of figures measure different products and windows, so adding them or computing a single CEX-versus-DeFi total would be misleading.
The Chaos Labs report also shows where the strain landed. Bitcoin fell from 116,500 to 102,700 dollars in about 30 minutes, some price updates were delayed by 5 blocks on certain markets, and Aave accrued about 400,000 dollars of deficits, mostly on CRV and ENS debt positions, around 500,000 dollars including expected bad debt. Against that, the report estimates a net profit of 1.5 million dollars for the protocol from fees and oracle revenue. The lending side bent without breaking; the cascade was mostly on leveraged trading venues.
A record week in February 2026 and a liquidation without a crash
In the selloff from 31 January to 5 February 2026, Aave processed 429 million dollars of liquidations in about 12,500 transactions, more than the 362 million across over 5,500 events of May 2021 on Aave V2. Aave’s own account of its history adds a detail about where liquidations happen: counted by number of events across all its markets since launch, 80 percent took place on networks other than Ethereum, Polygon alone close to half (46 percent). Measured in dollars, Ethereum dominates. Cheap chains make small positions worth liquidating.
Liquidations can also fire when nothing crashes. On 10 March 2026 a rate limit in Aave’s price adapter for wstETH, which by design lets the recorded exchange rate rise by at most 3 percent per three-day window according to Chaos Labs’ post-mortem, left the token’s exchange rate about 2.85 percent below its real value. That was enough to push eMode positions over the line: about 10,938 wstETH were liquidated across 34 accounts. The post-mortem states that no bad debt accrued; part of the liquidators’ gains was recovered, and the DAO planned to compensate the rest, no more than 358 ETH. The borrowers were liquidated by a configuration, not by the market.
The incidents of this lesson in one table, with the figures as reported by each source.
| Date | Event | Figure | Source |
|---|---|---|---|
| 12 March 2020 | Maker Black Thursday, zero-dollar bids | surplus from about +0.5 million to about -4 million dollars | Maker forum |
| 19-28 March 2020 | MKR debt auctions | 20,980 MKR sold for 5.3 million DAI | The Block |
| 22 November 2022 | CRV position liquidated on Aave V2 | about 1.5 to 1.6 million dollars of bad debt (about 2.6 million CRV) | Aave forum, Decrypt, Kaiko |
| 13 June 2024 | Curve founder’s CRV loans liquidated on five protocols | about 141 million dollars of CRV; more than 1 million of Llamalend bad debt | Decrypt citing Arkham |
| 10-11 October 2025 | Leveraged positions on derivatives venues | more than 19 billion dollars in 24 hours | CCN |
| 10-11 October 2025 | Aave liquidations (lending, separate product) | about 180 million dollars; deficit about 400,000 | Chaos Labs |
| 31 January-5 February 2026 | Aave liquidations in a selloff | 429 million dollars, about 12,500 transactions | Aave blog |
| 10 March 2026 | wstETH rate cap error on Aave | about 10,938 wstETH liquidated, 34 accounts, no bad debt | Chaos Labs post-mortem |
Protocol defences: caps, isolation mode, Umbrella and better oracles
Since 2022 lending protocols have added layers meant to keep a bad day from turning into bad debt. None removes the risk; each narrows one path to it.
- Supply and borrow caps limit how much of an asset the protocol accepts as collateral or lends out, so a single thin token cannot grow larger than the market can absorb in a liquidation.
- Isolation mode, which Aave V3 used until v3.7, let a riskier asset back loans only in a set basket of stablecoins up to a debt ceiling, and siloed borrowing kept the borrowers of some assets from holding other debts. The v3.7.0 release of 5 August 2026 removed both from the code and added an isolated flag for eMode categories instead, so older tutorials describe tools the current pool no longer has.
- Higher liquidation thresholds and bonuses for liquid assets, lower thresholds and higher bonuses for volatile ones, plus freezing markets that become too thin, as Aave did with CRV.
- The close factor and minimum leftover rules on Aave, which stop a liquidator from taking more than needed from a mildly unhealthy account while preventing dust that nobody would clear.
- A reserve factor, the share of borrowers’ interest kept by the protocol, which feeds the treasury that backs the last line of defence; on Aave it comes after Umbrella’s staked cover.
- Oracle design. Since March 2025 Aave’s Core Ethereum market has used Chainlink SVR, which auctions the right to backrun the price updates that open liquidations; Aave’s blog says it handled 675 million dollars of liquidations in its first nine months to early February 2026 and recaptured about 16 million dollars, split 65 percent to Aave and 35 percent to Chainlink. Rate caps on exchange-rate oracles limit manipulation, at the price of errors like the one of March 2026.
The limits are visible in the incidents above. The CRV position of 2022 sat on Aave V2, which had no supply or borrow caps; those arrived with V3. And a protective rate cap caused the March 2026 liquidations. Defences are trade-offs between protecting lenders and charging borrowers, set by governance votes anyone can read.
How to read a liquidation on a block explorer
Every liquidation leaves a public record. On Aave V3, open the transaction on Etherscan or the explorer of the network and look at the Logs tab for an event named LiquidationCall. It lists the collateral asset, the debt asset, the borrower’s address, debtToCover (the amount repaid), liquidatedCollateralAmount (what the liquidator took), the liquidator’s address and receiveAToken, which says whether the liquidator took aTokens or the underlying asset. Dividing the collateral taken by the debt repaid, at the oracle prices of that block, gives the bonus actually paid.
Three more checks help. A flash loan event in the same transaction shows a capital-free liquidation. A swap event right after it shows the collateral being sold at once, which is the selling pressure of a cascade. A DeficitCreated event means the account was emptied and debt was left behind. On Compound III the events are AbsorbCollateral and BuyCollateral; on Sky, the Clipper contract emits Kick and Take for each auction.
Habits that keep a loan away from liquidators
This is general education, not advice on any position. Most liquidations follow from a small buffer, not from bad luck. The previous lesson showed how to compute your liquidation price; the habit is to know it, keep the health factor well above 1 and set an alert well before the line rather than at it.
Watch the collateral more than the debt. A stablecoin debt against volatile collateral can be liquidated in a single sharp move, and thin tokens carry higher bonuses, which make a liquidation costlier when it comes. On Aave the first liquidation can take half of the debt with its bonus, and a second can follow minutes later. Partial repayment or added collateral before that point costs only gas; after it, the bonus is gone.
The March 2026 case carries its own warning: eMode and leverage loops run close to the threshold on purpose, which leaves no room for an oracle error. The next lesson looks at those borrowing strategies, from raising cash without selling to loops, and their costs.
