Soft Liquidations: How Curve LLAMMA and crvUSD Replace the Liquidation Cliff

Almost every lending protocol handles a falling collateral price the same way. There is a threshold, nothing happens above it, and below it a liquidator seizes collateral and takes a penalty. The design is simple, it is enforceable, and it produces a specific pathology: a small price move across an arbitrary line converts a solvent position into a large realized loss, and does so at the moment liquidity is worst.

Curve's approach to crvUSD lending replaces the line with a range. Instead of one price at which everything happens, there is a band of prices across which your collateral is progressively converted, and progressively converted back if the price recovers.

The result is not a free lunch, and understanding why is the entire point of this article. Soft liquidation does not remove the cost of a falling market. It changes the cost from a low-probability large hit into a high-probability small drain, and whether that is an improvement depends on something specific about the market you expect.

1. What LLAMMA actually is

Curve's developer documentation describes LLAMMA as "a market-making algorithm that rebalances the collateral of a loan using an AMM." The key structural difference from a conventional lending market is stated directly: rather than a single liquidation price, LLAMMA has "a liquidation range in which collateral is continuously being rebalanced."

Bands

The mechanism borrows the concentrated liquidity idea from Uniswap V3. When a loan is created, the borrower's collateral is distributed equally across a chosen number of price bands, with a minimum of 4 and a maximum of 50. Each band has an upper and lower oracle price boundary, and together the bands define the total liquidation range. Outside that range, no liquidation occurs at all.

At any moment each band is in one of three states, per the AMM documentation:

Nothing forces the conversion

This is the part that surprises people. There is no liquidation bot repaying debt and seizing collateral. LLAMMA maintains two prices: price_oracle, taken from an external oracle, and get_p, the AMM's own internal price, which depends on the balances in the active band and the oracle price. When those two diverge, ordinary arbitrage traders profit by trading them back together, and that trading is the rebalancing.

Soft liquidation is therefore self-executing through market incentives rather than through a privileged liquidation transaction. Curve's documentation summarizes the direction plainly: "The further down the price of the collateral goes, the more ETH will be sold for crvUSD."

2. De-liquidation and the round trip

When the price recovers, the process runs backwards. Curve's documentation describes de-liquidation as "converting the crvUSD backing the loan obtained through earlier soft-liquidation back into the collateral token (ETH) again."

So a borrower whose collateral dipped into the range and then recovered ends up back in collateral. Not all of it, though, and the shortfall is where the real economics of this design live.

A worked round trip

Deposit 10 ETH at $3,000, worth $30,000, and borrow $18,000 crvUSD. Suppose the chosen bands place the liquidation range roughly between $2,300 and $1,900.

Trace the arithmetic. Roughly 5 ETH was sold on the way down at an average of about $2,100, raising about $10,500. On the way back up, de-liquidation repurchases ETH at a higher average of about $2,300, so that $10,500 buys back about 4.57 ETH rather than the 5 that were sold.

Those two averages are the important part, and they are worth being precise about, because they are not the midpoints of the range. Selling evenly from $2,300 down to $2,100 would average $2,200, and buying evenly back would average $2,200 as well — a perfectly symmetric round trip returns all 5 ETH and costs nothing. The gap between the two averages is the whole loss, and it exists because LLAMMA's internal price is deliberately offset from the oracle. Arbitrageurs are the counterparty: they buy the collateral from the AMM a little below the oracle price on the way down and sell it back a little above on the way up, and the spread they capture is what leaves the borrower's position. The figures above stand in for that spread rather than being derived from the band boundaries.

You end with about 9.57 ETH instead of 10. The price finished exactly where it was when you entered the range, and the position is down about 4.3%.

Figure 1 turns that same round trip into something you can drag through directly, rather than read off a table.

Reading Figure 1

The figure opens exactly where the worked example's third row sits: price at $2,100, about 5 ETH and about $10,500 crvUSD. From there, put it through the same move the table above describes, in the order that reproduces it.

1. Drag the slider up to $2,500 first. This is the recovery leg from the table. ETH held climbs, but only to about 9.57, not back to 10. crvUSD held drops to zero, since the position is fully de-liquidated once the price clears the top of the range. The loss readout turns to clay and settles at about −4.3%, matching the recovery row above exactly.
2. Press Reset, then drag down toward $1,900 instead. Bands turn from green to copper one at a time as the marker crosses them, the ETH-held readout falls, and the crvUSD-held readout rises. Watch the "vs. holding 10 ETH" readout while you do this: it stays at 0.0% the whole way down. Nothing has been lost yet, because the crvUSD raised at each point is marked at the price it was raised at, not compared against a higher price the way the recovery leg is.
3. Notice where the loss actually appears. It is not visible while the price is falling, in either direction you try it. It shows up specifically once de-liquidation has to repurchase collateral at a higher price than the price it sold at. That gap between the sale price and the repurchase price is the entire cost, made visible rather than described. Reset again and drag all the way down to $1,900 before bringing it back up, and the same shape shows up at a larger scale: a full traverse of the range costs proportionally more than the half traverse the worked example uses, because there is more of the sell-low, buy-high gap to pay for.

The bands convert one at a time, in the order the price reaches them. The band nearest $2,300 is the first to turn from collateral to crvUSD, and the band nearest $1,900 is the last. That ordering is what "the active band" in the table in section 1 refers to: at any moment exactly one band is mixed, everything above it is still entirely collateral, and everything below it is already entirely crvUSD.

A price that never recovers past its own low point shows no loss in the figure, and that is the honest answer, not a bug. The realized cost only exists once de-liquidation actually buys back at a worse price than the sale. A position that dips into the range and is closed out for other reasons while still near its low has not yet paid the round-trip cost that this figure and the table above describe — it pays it precisely when the price comes back. That is also why the slider only ratchets its low-water mark downward rather than resetting on every drag: Reset is there so you can return to the example's own starting point on purpose, not because the position itself gets to forget how far it fell.

What this example actually tells you

The loss comes from the direction of the trades, not from a fee. LLAMMA sells collateral as the price falls and buys it back as the price rises. That is selling low and buying high, executed systematically. It is the same mechanical cost that liquidity providers face in a spot AMM, which is covered in impermanent loss explained, applied here to a borrower's collateral rather than to a liquidity position.

The cost is incurred in both directions. Curve's documentation is explicit that rebalancing losses accrue while the loan sits inside the liquidation range, on the way down and on the way up. Recovery does not refund the drawdown. It completes the round trip that produced the loss.

And it is charged for a service that may never have been needed. In this example the price never approached insolvency. A conventional lending market with the same loan-to-value would have charged nothing at all for this move, because the liquidation threshold was never crossed. The 4.3% is the premium paid for protection against a scenario that did not happen.

3. Why a soft-liquidated position is short volatility

Put the last two sections together and the payoff structure becomes clear.

A borrower in a conventional market holds something option-like. Most of the time nothing happens. Occasionally the price crosses the threshold and they take a single large, discrete hit. Their cost is zero until it is large.

A borrower inside a LLAMMA range holds the opposite. There is no threshold event to fear, and there is a continuous drain any time the price moves back and forth through their bands. Their cost is small and constant, and it accumulates with realized volatility.

That makes a soft-liquidated position structurally short volatility. Every oscillation through the range costs something, and it is worth being concrete about where it goes: to the arbitrageurs who take the other side of each conversion, as described in the round trip above. The borrower is not paying a protocol fee. They are paying the spread that makes arbitrage worth doing, which is also what makes the mechanism function at all. A market that chops sideways inside the range for weeks charges the borrower repeatedly for a liquidation that never happens.

This also explains a behavior that confuses new crvUSD borrowers: the position keeps losing value even as the price recovers. Nothing is broken. The mechanism is doing exactly what it is designed to do, and the borrower is paying the premium on the insurance they bought.

4. Choosing the band count

Band count is the main lever a borrower controls, and Curve's documentation states the trade-off directly: a greater number of bands makes liquidation more gradual but starts it earlier as the price falls, while fewer bands start the process later but make it more abrupt. It also notes that more bands and more liquidity reduce losses.

Those two effects point in opposite directions, which is what makes the choice interesting.

What this trade-off actually tells you

More bands lower the rate of loss but raise the exposure to it. Spreading collateral across 50 bands makes each individual conversion cheaper, and it also widens the price range in which you are paying anything at all. If the price spends most of its time oscillating inside a wide range, a low loss rate applied over far more time can exceed a high loss rate applied briefly.

The question is not "how gently do I want to be liquidated." It is "how much of the price distribution do I want to spend inside the paying zone." A borrower who is confident the price will hold well above the range should prefer a narrow range they rarely touch. A borrower who expects to spend real time inside it should prefer the gentler rate.

Curve's Controller also offers a direct health buffer. Its V3 Controller documentation describes a user-configurable extra_health parameter that adds a health buffer when entering soft liquidation. That is a separate lever from band count, trading borrowing capacity for a later start to the rebalancing cost.

5. Where soft liquidation still fails

Soft liquidation is not an escape from liquidation. It is a different shape of it, and three failure modes survive.

Hard liquidation still exists. If health deteriorates far enough, the loan is force-closed and the position ends. Soft liquidation delays and softens the path toward that point, but it does not remove the endpoint. A sufficiently deep and sustained decline exhausts the range, and past that the mechanism has nothing left to convert.

The rebalancing depends on arbitrageurs showing up. Because conversion is executed by traders equalizing the oracle price with the AMM price, its quality depends on arbitrage being profitable and executable. Curve's documentation lists arbitrage efficiency alongside band count and available liquidity as a determinant of rebalancing losses. During a gas spike, a chain congestion event, or a violent gap, the arbitrage may lag, meaning the position is under-rebalanced at exactly the moment rebalancing matters most.

Oracle behavior drives everything. Band boundaries are defined in oracle prices and the active band is determined by the oracle. A stale, laggy, or manipulated feed moves the entire rebalancing schedule relative to the real market. Soft liquidation removes the cliff but keeps the oracle dependency common to every collateralized lending design.

6. Soft versus hard liquidation compared

7. Conclusion

LLAMMA is one of the more genuinely original mechanisms in DeFi lending, and it is frequently described in a way that overstates what it does. It does not make liquidation cheap or optional. It replaces a discrete, penalty-driven event with a continuous, market-driven rebalancing, and the cost profile changes accordingly.

The mechanics are straightforward once the framing is right. Collateral is spread across 4 to 50 bands. As the oracle price falls through them, arbitrageurs convert collateral into crvUSD because doing so is profitable. As the price recovers, the same arbitrage converts it back. The round trip sells low and buys high by construction, so the position loses value even when the price ends where it started, and interest accrues throughout regardless.

That produces a clear decision rule. If your fear is a violent one-directional gap through a liquidation price, soft liquidation is real protection and worth paying for. If your expectation is a choppy market that repeatedly touches your range without ever threatening solvency, you are buying insurance against an event that will not occur while paying the premium every time the price moves. In that case a conventional market with a more conservative loan-to-value ratio is cheaper.

The lever you control is band count, and the right way to think about it is not how gently you want to be liquidated but how much of the price distribution you expect to spend inside the zone where you are paying at all.

For how this compares with other leveraged strategies, see advanced DeFi yield and hedging strategies. For the conventional liquidation designs this is reacting against, see leveraged yield and credit accounts and crypto liquidations explained.

Frequently asked questions

Sources and further reading

Primary protocol documentation:

• Curve Technical Docs — LLAMMA explainer
• Curve Technical Docs — crvUSD AMM and bands
• Curve Technical Docs — crvUSD Controller

Related CoinBeaver articles:

• Advanced DeFi yield and hedging strategies
• Leveraged yield and credit accounts: Gearbox vs Morpho Blue
• EigenLayer, Karak, and LRTs: the restaking risk loop
• Crypto liquidations explained
• Impermanent loss explained
• DeFi lending explained

This article is educational and is not financial advice. All prices, band ranges, position sizes, and loss percentages above are illustrative examples constructed to show the mechanism, not live market data or protocol quotes. Actual band boundaries depend on the market's parameters and the number of bands chosen, and realized rebalancing losses depend on volatility, available liquidity, and arbitrage conditions. Verify current market parameters in Curve's own documentation and application before borrowing.