Skip to content
DeFi Reference

CDP-backed stablecoins — DAI/USDS, LUSD, crvUSD

1. TL;DR

Section titled “1. TL;DR”

A collateralized-debt-position (CDP) stablecoin lets you lock crypto worth more than the dollars you want, and mints those dollars against it; if the collateral’s value falls too close to the debt, the system sells enough of it to make the protocol whole again. MakerDAO’s DAI (now largely rebranded USDS under Sky) pioneered this on Ethereum in 2017, later adding a Peg Stability Module (PSM) that swaps DAI 1:1 against USDC and a Dai Savings Rate (DSR) that pays holders for locking DAI. Curve’s crvUSD replaced discrete liquidation auctions with LLAMMA, an AMM that continuously and gradually converts collateral to crvUSD as price falls, while Liquity’s LUSD (and its V2 successor BOLD) minimizes governance and lets users set their own interest rates.

2. Explain it simply

Section titled “2. Explain it simply”

Analogy

Section titled “Analogy”

It’s a pawnshop for crypto: you hand over an item worth more than the cash you want (say a watch worth $1,000 for a $600 loan), and the shop keeps the watch as long as your loan is outstanding. If the watch’s resale value later drops close to $600, the shop doesn’t wait for you to default — it sells the watch immediately to recover its $600 plus a fee, and would hand you back anything left over. The “shop” here is a smart contract instead of a person, and it can check the watch’s value continuously instead of waiting for you to walk in.

ELI5

Section titled “ELI5”

Imagine you own a valuable trading card worth $100 and you want $60 in spending money without selling it. A vault lets you lock the card and get $60 in a new kind of dollar. If your card’s value later drops to $65, the vault worries it won’t get its $60 back, so it automatically sells the card, pays itself back, and gives you whatever’s left. As long as more people want to lock cards than take their money back out, this new dollar keeps circulating and can be traded like any other.

Step-by-step walkthrough

Section titled “Step-by-step walkthrough”

Scenario: Alice opens a vault with ETH collateral to mint DAI/USDS.

  1. Before. Alice holds 10 ETH, worth $3,000 each ($30,000 total), and 0 DAI. The vault type requires a 150% liquidation ratio.
  2. Deposit and mint. Alice locks her 10 ETH in the vault and mints 15,000 DAI, giving her a starting collateral ratio of 30,000/15,000=200%30{,}000/15{,}000 = 200\% — a safety cushion above the 150% liquidation line. State: Alice holds 15,000 DAI and 0 free ETH; the vault holds 10 ETH backing 15,000 DAI of debt.
  3. Fee accrues. Over time a stability fee (interest on the minted debt, conceptually the same continuous-rate accrual as /lending/interest-rate-models/) compounds into Alice’s debt; suppose it grows to 15,750 DAI owed.
  4. Price falls. ETH drops to $2,363. Collateral value is now $23{,}630,and, and 23{,}630/15{,}750 = 150%$ — exactly the liquidation ratio.
  5. Liquidation. The vault is flagged; collateral is auctioned (or, in crvUSD’s design, continuously converted — see §3) to repay the 15,750 DAI debt plus a liquidation penalty. After the auction/penalty, any leftover collateral value is returned to Alice; she keeps the 15,000 DAI she originally minted but loses her ETH position and pays the penalty.

Common misconceptions

Section titled “Common misconceptions”
  • Myth: DAI is basically the same as USDC. Reality: DAI/USDS is primarily backed by overcollateralized crypto and real-world-asset vaults, though the PSM also lets USDC make up part of the backing (see §3 and /stablecoins/pegs-and-psm/).
  • Myth: Liquidation means losing all your collateral. Reality: collateral is sold only to cover the debt plus a penalty; any surplus value is returned to the vault owner.
  • Myth: Holding DAI automatically earns the Dai Savings Rate. Reality: a holder must actively lock DAI into the Pot contract via a join transaction — “the savings rate is not natively integrated into the ERC20 DAI token” (Sky Protocol docs).
  • Myth: crvUSD’s LLAMMA liquidates everything at once when collateral price drops. Reality: soft liquidation converts collateral to crvUSD gradually, band by band, as the price crosses each band’s range — not in one discrete event.
  • Myth: Overcollateralization makes CDP stablecoins risk-free. Reality: a fast enough price crash combined with network congestion can outrun liquidations entirely, as happened on Black Thursday (see §6).

If you only remember one thing

Section titled “If you only remember one thing”

A CDP stablecoin is a loan against crypto you keep the upside on until the price falls too far, at which point the contract sells your collateral automatically to make itself whole.

3. How it works

Section titled “3. How it works”

Collateral ratio and liquidation

Section titled “Collateral ratio and liquidation”

Define collateral value V=Pc⋅QV = P_c \cdot Q (price of collateral times quantity locked) and outstanding debt DD (principal plus accrued stability fee). The collateral ratio is

CR=VDCR = \frac{V}{D}

Each vault type sets a liquidation ratio CRliqCR_{\text{liq}} (the minimum CRCR before liquidation is triggered) and, at the point of liquidation, a penalty ϕ\phi is added to the debt the borrower must cover. Maker/Sky uses discrete liquidation: at CR<CRliqCR < CR_{\text{liq}} the vault’s full collateral goes to a Dutch/English-style auction (mechanics shared with general lending liquidations at /lending/liquidations/) to raise D(1+ϕ)D(1+\phi); any auction proceeds above that are returned to the vault owner.

The Peg Stability Module (PSM)

Section titled “The Peg Stability Module (PSM)”

The PSM lets anyone swap USDC for DAI/USDS (and back) at a fixed rate rather than opening a vault. Per Sky’s own documentation, “LitePSMs stabilize the USDS and DAI stablecoins against their pegged value by enabling users to trade USDS or DAI with other stablecoins like USDC (referred to as ‘gems’) at fixed exchange ratios and fees,” through two functions: sellGem (“Sell USDC and receive USDS”) and buyGem (“Buy USDC with USDS”) (Sky Protocol docs, fetched 2026-08-29). Fees tin (on sellGem) and tout (on buyGem) are governance-set — “a tin value of 0.01 … would assess a fee of 1% (100 basis points)” — and governance can halt either function by setting the fee to the maximum uint value. A buf parameter caps “the maximum DAI balance held by LitePSM-DAI-USDC at any time” (Sky Protocol docs, fetched 2026-08-29). This makes the PSM the same no-arbitrage mechanism as §3 of /stablecoins/fiat-backed-and-cctp/, except it runs on-chain against DAI/USDS’s own reserve rather than an off-chain issuer redemption desk — with the trade-off that DAI’s peg becomes partially dependent on USDC’s own solvency (see §6).

The Dai Savings Rate (DSR)

Section titled “The Dai Savings Rate (DSR)”

The DSR pays DAI holders for locking DAI in the Pot contract, funded by stability-fee revenue: “the Pot contract generates additional dai by recording bad debt first in Vow which will most likely get cancelled with surplus that is also being collected continuously as stability fees from Vaults” (Sky Protocol docs, fetched 2026-08-29). Balances compound via a rate accumulator: “normalized balances of users are stored in the pie variable and the real dai balance of users can be calculated by multiplying pie and chi anytime,” where chi updates on each drip call. The DSR (now the Sky Savings Rate for USDS) is a deliberate demand lever: “governance uses this rate to increase demand by making it attractive for users to purchase and hold dai” (Sky Protocol docs, fetched 2026-08-29).

crvUSD and LLAMMA

Section titled “crvUSD and LLAMMA”

Curve’s crvUSD replaces discrete liquidation with LLAMMA (Lending-Liquidating AMM Algorithm), described by Curve’s docs as “the market-making contract that rebalances the collateral of a loan,” implemented as “an algorithm implemented into a smart contract which is responsible for liquidating and de-liquidating collateral based on market conditions” (Curve docs, fetched 2026-08-29). Collateral for a loan is spread across 4–50 price bands (structurally similar to Uniswap v3 ticks), each with an upper/lower oracle price. As the collateral price falls into a band, LLAMMA sells a slice of that band’s collateral for crvUSD via arbitrageurs — soft liquidation — and as price recovers, arbitrageurs convert crvUSD back to collateral in that band — de-liquidation — described in the docs as: “when collateral price increases, the system performs ‘de-liquidation’… conversely, when collateral price falls, ‘soft-liquidation’ converts collateral into crvUSD through arbitrage… The process occurs gradually across multiple bands rather than immediate liquidation” (Curve docs, fetched 2026-08-29). Because this relies on arbitrageurs finding it profitable to trade against each band, LLAMMA intentionally prices bands aggressively relative to the oracle so the arbitrage stays attractive.

Liquity’s LUSD and V2’s BOLD

Section titled “Liquity’s LUSD and V2’s BOLD”

Liquity’s original LUSD minimizes governance and oracle dependence by using a Stability Pool that absorbs liquidated debt directly rather than running an auction. Liquity V2 introduces BOLD, described in its whitepaper as offering “user-set interest rates and the ability to borrow against multiple collateral assets,” including ETH and liquid staking tokens (Lido’s wstETH and Rocket Pool’s rETH), plus “a novel adaptive redemption mechanism” that lets the market — not governance — balance borrowers and stability seekers by redeeming BOLD against the lowest-interest-rate troves first, incentivizing borrowers to price their own risk (Liquity V2 whitepaper, 2025-01-28).

4. Worked numeric example

Section titled “4. Worked numeric example”

Continue Alice’s vault from §2, now contrasting Maker/Sky’s discrete auction against crvUSD’s continuous LLAMMA over the same ETH price path ($3,000 → $2,363, a 21.2% drop):

Maker/Sky-style discrete liquidation:

  • Debt at liquidation: D=15,750D = 15{,}750 DAI; collateral value at trigger: V=23,630V = 23{,}630; liquidation penalty ϕ=13%\phi = 13\% (illustrative, a typical order of magnitude for ETH vaults in Maker governance parameter votes).
  • Amount owed after penalty: D(1+ϕ)=15,750×1.13=17,797.5D(1+\phi) = 15{,}750 \times 1.13 = 17{,}797.5 DAI.
  • Auction sells enough ETH to raise 17,797.5 DAI; at the liquidation-trigger price of $2,363/ETH that is $17{,}797.5 / 2{,}363 \approx 7.53$ ETH.
  • Alice’s remaining collateral: 10−7.53=2.4710 - 7.53 = 2.47 ETH (worth 5,8385{,}838 at $2,363), returned to her. Net result: she keeps her original 15,000 DAI, loses 7.53 ETH, and receives 2.47 ETH back — a one-time, all-at-once event.

crvUSD-style LLAMMA (illustrative): suppose Alice’s 10 ETH is spread across 10 bands from $2,900 down to $2,300. As price falls from $3,000 to $2,363, it passes through roughly 6 of those bands, so LLAMMA has already converted about 60% of her collateral (6 ETH) into crvUSD by the time price reaches $2,363, leaving 4 ETH still as collateral — no discrete penalty event, no auction, and if price recovers to $2,900 before she acts, de-liquidation converts crvUSD back into ETH automatically. The trade-off: Alice’s collateral composition is now uncertain (partly ETH, partly crvUSD) rather than fixed until a single liquidation event.

Both mechanisms move roughly similar economic value out of Alice’s position as price falls the same amount, but Maker’s is a single sharp event with a fixed penalty, while LLAMMA’s is continuous and reversible.

5. Where it’s used

Section titled “5. Where it’s used”

Ethereum

Section titled “Ethereum”
  • Sky (formerly MakerDAO) — DAI/USDS — the largest CDP stablecoin; vaults accept ETH, staked ETH, and real-world assets, plus a PSM leg backed by USDC. developers.skyeco.com
  • PSM — DAI/USDS ↔ USDC swaps at a fixed rate via sellGem/buyGem; see /stablecoins/pegs-and-psm/ for the full peg-mechanics treatment.
  • DSR / Sky Savings Rate — native yield for locking DAI/USDS, funded by stability-fee revenue.
  • Curve crvUSD — LLAMMA-based soft-liquidation stablecoin minted against ETH, staked ETH, and other Curve-listed collateral. docs.curve.finance
  • Liquity LUSD / BOLD (V2) — minimal-governance, Stability-Pool-based liquidation; V2 adds user-set interest rates and multi-collateral support (ETH, wstETH, rETH). liquity.gitbook.io

Solana

Section titled “Solana”

6. Risks, attacks, and incidents

Section titled “6. Risks, attacks, and incidents”
  • Black Thursday, MakerDAO, 12 March 2020. ETH fell roughly 43% in hours; Ethereum gas prices spiked as every liquidation keeper tried to bid simultaneously, and some keeper bots with stuck transactions were absent from auctions entirely, letting a small number of bidders win collateral auctions with $0 DAI bids — about $8.32 million of ETH was won for free in one estimate. The shortfall left MakerDAO with bad debt estimated between $4.5 million and $5.67 million (sources vary), which the protocol covered by auctioning newly minted MKR for DAI (widely reported post-mortems, 2020-03; e.g. CryptoSlate, Medium/Linum Labs analyses). Root cause: liquidation auctions assumed competitive bidding under normal gas conditions, which failed under simultaneous network congestion.
  • PSM concentration risk realized indirectly, March 2023. When USDC depegged after Circle’s Silicon Valley Bank exposure (see /stablecoins/fiat-backed-and-cctp/ and /stablecoins/pegs-and-psm/), USDC represented roughly 51.87% of DAI’s collateral ($4.42 billion) at the time via the PSM, and DAI itself lost about 7.4% of its value (CoinDesk/Cointelegraph contemporaneous reporting, 2023-03). MakerDAO governance responded within days by raising PSM swap fees and capping the daily USDC-to-DAI mint. Root cause: a fixed-rate PSM transmits its collateral’s depeg directly into DAI’s price, proportional to the PSM’s share of total backing.
  • Oracle and arbitrage dependency in LLAMMA. Soft liquidation depends on arbitrageurs finding band trades profitable; if arbitrage infrastructure fails or an oracle misreports price during extreme volatility, bands may not rebalance as designed (mechanism-level risk noted in Curve’s own documentation of the arbitrage-driven design; no crvUSD-specific large loss event has been reported in these sources as of 2026-08).
  • Redemption pressure under Liquity V2’s adaptive mechanism. Borrowers who set interest rates too low relative to the market risk having their troves redeemed against first, since BOLD redemptions target the lowest-rate troves — a market-driven risk transfer rather than a bug, but one that changes the calculus of picking a rate (Liquity V2 whitepaper, 2025-01-28).

7. Open problems

Section titled “7. Open problems”
  • How much of DAI/USDS should be USDC-backed? The PSM’s convenience for arbitrage trades directly against its concentration risk; MakerDAO/Sky governance has repeatedly adjusted PSM fees and caps rather than settling this trade-off permanently (see /stablecoins/pegs-and-psm/).
  • Auction design under network stress. Black Thursday exposed that liquidation auctions assuming competitive bidding can fail exactly when they are needed most; Sky’s current auction and keeper incentive design continues to evolve in response (see /lending/liquidations/).
  • Continuous vs. discrete liquidation. Whether LLAMMA’s gradual, reversible approach or Maker’s discrete-and-final approach better protects borrowers and the protocol under different volatility regimes is an active design debate without a settled answer in these sources.
  • Governance minimization vs. adaptability. Liquity’s immutable, minimal-governance design (see also /stablecoins/algorithmic-failures/ for MetaCoin’s similar bet) trades away the ability to intervene the way Sky did after Black Thursday and the SVB episode — whether that trade-off is worth it is contested rather than resolved.

8. Ethereum vs Solana

Section titled “8. Ethereum vs Solana”
AspectEthereumSolana
Flagship CDP stablecoinDAI/USDS (Sky), crvUSD (Curve)n/a at comparable scale
Liquidation styleDiscrete auctions (Sky) or continuous soft liquidation (crvUSD LLAMMA)n/a
Peg support beyond overcollateralizationPSM (USDC 1:1 swap), DSR/Sky Savings Raten/a
Minimal-governance alternativeLiquity LUSD/BOLDn/a
Where Solana’s dollar liquidity comes from instead—Fiat-backed issuers (USDC, USDT, PYUSD) and delta-neutral designs (USDe); see /stablecoins/fiat-backed-and-cctp/, /stablecoins/delta-neutral/

CDP stablecoins are a distinctly Ethereum-native pattern: they need a deep base of volatile collateral, mature liquidation/keeper infrastructure, and (for crvUSD) an established AMM ecosystem to build LLAMMA on top of — all things Ethereum accumulated over years before Solana’s DeFi stack matured. Solana’s faster, cheaper execution reduces the Black-Thursday-style congestion risk in principle, but no protocol there has attracted the collateral base needed to make a CDP stablecoin systemically relevant; Solana’s dollar liquidity took the fiat-backed and delta-neutral paths instead.

9. Reference doc

Section titled “9. Reference doc”

The reference

Section titled “The reference”

Rate Mechanism (Dai Savings Rate) and LitePSM — Sky Protocol Docs, fetched 2026-08-29. developers.skyeco.com/deep-dives/rate-mechanism and developers.skyeco.com/guides/psm/litepsm

Summary of the reference

Section titled “Summary of the reference”

These two Sky Protocol documentation pages describe the two peg-support mechanisms layered on top of Sky’s core overcollateralized-vault system. The Rate Mechanism page explains the Dai Savings Rate (DSR): it is funded entirely from stability-fee revenue collected on vaults, flowing through the Vow (surplus/deficit buffer) into the Pot contract. Unlike a passive bank interest rate, users must actively join the Pot to start earning — the docs stress that “the savings rate is not natively integrated into the ERC20 DAI token.” Balances are tracked as normalized units (pie) multiplied by a continuously updating rate accumulator (chi), which increases whenever anyone calls drip. The stated policy purpose is demand management: governance raises the DSR to make holding DAI more attractive when the peg needs support from the demand side, complementing the stability fee’s supply-side lever on borrowers.

The LitePSM page documents the peg’s other lever: a gas-optimized Peg Stability Module that lets anyone swap USDC for USDS/DAI or back at a fixed rate via sellGem/buyGem. Fees (tin on sells, tout on buys) are set in WAD-scaled fixed point by governance and can be set to a maximum value to halt either direction entirely — a circuit breaker Sky used during the March 2023 USDC depeg (see §6). A buf parameter caps the DAI balance the LitePSM can hold at any time, bounding how much of DAI’s supply can be backed through this single, concentrated USDC channel.

Together, the two pages show DAI/USDS’s peg is maintained by three coordinated levers: the stability fee (cost of new debt), the DSR/Savings Rate (reward for holding), and the PSM (a hard arbitrage bound against USDC) — none of which is sufficient alone, and all of which are governance-tunable rather than fixed in code.

Key quotes

Section titled “Key quotes”

“The savings rate is not natively integrated into the ERC20 DAI token.” (Rate Mechanism, Funding Source)

“Ethereum external accounts or smart contracts have to execute a join transaction in Pot to earn the savings rate on their dai.” (Rate Mechanism, How Users Earn DSR)

“Sky governance uses this rate to increase demand by making it attractive for users to purchase and hold dai.” (Rate Mechanism, Role in Peg Stability)

“LitePSMs stabilize the USDS and DAI stablecoins against their pegged value by enabling users to trade USDS or DAI with other stablecoins like USDC (referred to as ‘gems’) at fixed exchange ratios and fees.” (LitePSM, Core Purpose)

“Governance can set either tin or tout to the value of type(uint256).max, halting sellGem and buyGem transactions respectively.” (LitePSM, Fee Structure)

How to read the original

Section titled “How to read the original”

Background needed: what a vault/CDP is (locking collateral to mint debt) and what “stability fee” means as vault interest (see /lending/interest-rate-models/). Skip the WAD fixed-point arithmetic details on a first pass — the concept (fees expressed as fractions of 10^18) matters more than the exact scaling. The hardest part to read correctly is the interaction between DSR funding and stability fees: it is tempting to think of the DSR as “yield from nowhere,” but it is a direct transfer from borrowers (who pay the stability fee) to savers (who lock DAI in the Pot) mediated by governance’s rate-setting, not an external yield source.

What changed since

Section titled “What changed since”
  • Sky’s rebrand from MakerDAO (DAI → USDS, MKR → SKY, DSR → Sky Savings Rate) restructured naming across all these mechanisms without changing their core logic.
  • LitePSM is itself a 2024-era gas-optimization of the original PSM contract design; older PSM instances (dss-psm) remain deployed alongside it.
  • The March 2023 USDC depeg (see §6) is the clearest real-world stress test of the PSM’s fee/cap circuit breakers described in this doc, and led directly to the parameter changes Sky made that week.

Secondary references

Section titled “Secondary references”
  1. Liquity V2 Whitepaper (below) — read for the alternative, minimal-governance design philosophy.
  2. /stablecoins/pegs-and-psm/ — read for the full mechanics and incident history of the PSM as a peg-defense tool.
  3. /lending/liquidations/ — read for how Sky’s Dutch/English collateral auctions compare to other lending-protocol liquidation designs.

The reference

Section titled “The reference”

LLAMMA: Core Concepts — Curve Finance Docs, fetched 2026-08-29. docs.curve.finance/developer/crvusd/amm

Summary of the reference

Section titled “Summary of the reference”

This page documents LLAMMA (Lending-Liquidating AMM Algorithm), the mechanism underpinning Curve’s crvUSD. It defines LLAMMA as “the market-making contract that rebalances the collateral of a loan” and, more mechanically, “an algorithm implemented into a smart contract which is responsible for liquidating and de-liquidating collateral based on market conditions.” Each crvUSD market (e.g., ETH<>crvUSD) runs its own AMM holding both the collateral and crvUSD.

The core structural idea is bands: price ranges into which a loan’s collateral is deposited, structurally similar to Uniswap v3’s concentrated-liquidity ticks. A borrower’s collateral spreads evenly across a chosen number of bands (4 to 50), each bounded by p_oracle_up and p_oracle_down, with a band-width factor A controlling how narrow (and how much internal leverage) each band has. At any moment, bands above the current price are fully in crvUSD, bands below are fully in the collateral asset, and the single “active band” straddling the current price holds a mix of both.

The page explains the two-directional rebalancing that gives LLAMMA its name: as collateral price falls into a band, soft liquidation converts that band’s collateral into crvUSD via arbitrage; as price rises back through a band, de-liquidation converts crvUSD back into collateral. Both happen gradually, band by band, rather than as one discrete liquidation event — the explicit contrast the docs draw with traditional liquidation. The page closes by contrasting LLAMMA with ordinary Curve StableSwap pools (see /exchange/stable-pools/): where StableSwap holds static liquidity for trading between similarly priced assets, LLAMMA actively redistributes collateral and debt tokens across bands in response to price, using arbitrage instead of protocol-triggered liquidation calls.

Key quotes

Section titled “Key quotes”

“LLAMMA… is the market-making contract that rebalances the collateral of a loan.” (What is LLAMMA)

“An algorithm implemented into a smart contract which is responsible for liquidating and de-liquidating collateral based on market conditions.” (What is LLAMMA)

“A band is a range of prices into which liquidity is deposited.” (Price Bands and Ticks, Band Structure)

“The process occurs gradually across multiple bands rather than immediate liquidation, allowing borrowers to maintain positions through price volatility.” (Soft Liquidation Mechanism)

How to read the original

Section titled “How to read the original”

Background needed: Uniswap v3-style concentrated liquidity/ticks (see /exchange/concentrated-liquidity/) and the general idea of liquidation in overcollateralized lending (see /lending/liquidations/). Skip the exact band-width factor A calibration details on a first pass. The hardest part to read correctly is the “active band” concept: it is not that one band is special by design, but that whichever band currently straddles the oracle price is, by construction, the only one holding a mix of both assets at any instant — every other band is entirely one asset or the other.

What changed since

Section titled “What changed since”
  • crvUSD launched in 2023 after this LLAMMA design was finalized; Curve has since extended the same AMM to power Curve Lending (permissionless LLAMMA-based lending markets for arbitrary collateral, not just crvUSD’s own governance-listed assets).
  • Community risk analyses (e.g., LlamaRisk’s Curve Lending primer) have since examined LLAMMA’s behavior under thin-liquidity collateral, a stress case not covered in the core docs page.

Secondary references

Section titled “Secondary references”
  1. Sky Protocol Rate Mechanism / LitePSM docs (above) — read for the discrete-liquidation alternative crvUSD was designed to improve on.
  2. /exchange/stable-pools/ — read for Curve’s original StableSwap invariant that LLAMMA extends into a liquidation-aware AMM.
  3. /lending/liquidations/ — read for how soft liquidation compares with Dutch-auction and keeper-bot liquidation models elsewhere.

10. Sources

Section titled “10. Sources”