.jpg)
Curve has accidentally published the best dataset in DeFi on what liquidation actually costs borrowers. Its crvUSD lending markets logged 704 soft-liquidation instances across 602 unique addresses, and the median position spent 14.5 days half-converted, neither dead nor safe, while a quarter of positions sat in the danger zone 38.9 days or longer. On Aave or Compound, those positions would have been executed on day one. On Curve's LLAMMA engine, they mostly survived.
The mechanism is the story. Conventional money markets run binary liquidations: cross the threshold, lose the collateral, done. Curve's Lending-Liquidating Automated Market Maker Algorithm instead converts collateral into crvUSD continuously across a price band, and if the price recovers before the band's floor is breached, the conversion reverses and the borrower keeps their position. Of the 704 instances, 476, or 67.6%, were initiated in the first half of 2026, meaning the dataset stress-tests the design through a full drawdown cycle rather than a quiet quarter.
"Soft liquidation is not free, however. The data shows borrowers can still lose money through trading fees, conversions, rebalancing, interest and repeated price moves in both directions."
The Hidden Bill for Staying Alive
Here is the part the marketing omits. A borrower who lingers 38.9 days in soft liquidation pays conversion slippage, rebalancing losses, compounding borrow interest, and AMM fees on every whipsaw, in both directions, the market makes. For volatile collateral, that accumulated friction can approach what a single hard liquidation would have cost. The mercy is priced; LLAMMA just spreads the invoice over two weeks instead of one savage minute. What Curve has really proven is that liquidation risk is a spectrum, not a switch, and that the market's rush to grade DeFi risk is only beginning to price this dimension. Compare a cascade event like the Notional escrow exploit, where distressed positions get zero recovery options: Curve's median borrower had two weeks of optionality.
The tail risks survive the upgrade, though. Soft liquidation is not bankruptcy protection: break below the lowest LLAMMA band and the position triggers a terminal hard liquidation anyway. The conversion engine depends on AMM depth, so a market-wide liquidity crunch or a crvUSD de-peg degrades exactly the mechanism that makes it humane. And the dataset says nothing about what borrowers actually paid in aggregate, only how long they stayed exposed.
What It Means for On-Chain Credit
The institutional read is the one worth keeping. Institutions evaluating on-chain credit have always priced the cascade: one bad candle liquidates a cluster, dumps collateral into thin pools, and liquidates the next cluster. LLAMMA's data shows a working alternative, 14.5-day median buffers that smooth market impact and make position-level survival the default rather than the exception. That matters at Curve's scale, roughly $2.35 billion in deposits, about $1.4 billion in 30-day DEX volume, and around $46 million in active loans, generating approximately $1.3 million in monthly fees. It also matters for how DeFi fee revenue gets competed for: risk architecture is becoming the product.
Watch whether Aave or Compound answers with their own gradual-liquidation designs. The 14.5-day number gives them something to beat, and gives lenders a new question to ask before the next drawdown: does my protocol execute borrowers, or does it carry them?
