Note: when ETH drops 15% in an hour, hundreds of positions simultaneously cross the liquidation threshold. The first liquidator to call absorb captures discounted collateral assets. The second gets nothing. This is a competitive environment measured in milliseconds: Compound liquidations are captured by MEV bots with direct connections to builders and custom Rust implementations. Our team, with over 7 years of experience in DeFi and having developed over 40 trading and liquidation bots, builds solutions that work in this environment. Average profit per successful liquidation is $150–250, and gas savings compared to naive approaches reach 35–45%.
Compound v3: A New Approach to Liquidations
According to Compound v3 (Comet) documentation, liquidation occurs in two steps: absorb and buyCollateral.
Compound v3 (Comet) differs significantly from v2 in its liquidation model. In v2, there was liquidateBorrow — the liquidator repays the borrower's debt and receives their collateral with an 8–15% bonus. In v3, a two-step model is introduced: absorb and buyCollateral. Profit on v3 liquidations comes from the arbitrage between the purchase price from Comet and the market price. The bot must atomically: call buyCollateral, sell the obtained asset on a DEX, and return the base token. Flash loans from Aave or Uniswap v3 for financing are a standard pattern.
How to Identify Insolvent Positions in Real Time?
A position is insolvent when the borrowing capacity falls below the debt. Compound v3 provides isLiquidatable(address account) and getBorrowableOf(...). Polling hundreds of thousands of positions via eth_call is impractical. An effective approach is event-based monitoring: listen for Supply, Withdraw, Transfer events from Comet, update a local copy of states. When the collateral price drops (Chainlink AnswerUpdated event), recalculate the health factor for positions with that collateral. The data structure is a sorted set in Redis by health factor, allowing O(log n) lookup of positions closest to liquidation.
| Parameter |
Naive |
Optimized |
| Detection |
Polling every 12 sec |
WebSocket + event-based (3x faster) |
| Submission |
Public mempool |
Flashbots bundle |
| Gas price |
Fixed |
Dynamic (80th percentile + boost) |
| Execution |
EOA transaction |
Liquidation contract (1 tx) |
Bot Architecture
Position Monitoring
Monitoring Details
Two levels: The Graph subgraph for historical data, WebSocket subscription via ethers.js provider.on or viem watchContractEvent for real-time. The subgraph updates with a 1–3 block delay — sufficient for competitive liquidations. Chainlink price feeds via AggregatorV3Interface with updatedAt staleness checks.
Liquidator Smart Contract
Atomic liquidation via flash loan:
contract CompoundLiquidator {
function liquidate(
address comet,
address[] calldata accounts,
address collateralAsset,
uint baseAmount,
address flashLoanPool // Uniswap v3 pool
) external {
// 1. Flash loan base token from Uniswap v3
// 2. absorb(address(this), accounts)
// 3. buyCollateral(collateralAsset, minOut, baseAmount, address(this))
// 4. Swap collateral -> base token via DEX
// 5. Return flash loan + fee
// 6. Profit goes to msg.sender or treasury
}
}
A critical detail: absorb and buyCollateral are two separate calls. Between them, another bot could buy the collateral. You must either check available collateral before purchase (quoteCollateral) or accept that in rare cases the transaction reverts.
Profitability Filter
Not every liquidatable position is profitable. Before submitting a transaction, calculate:
profit = buyCollateralValue * (1 - storeFrontPriceFactor) - flashLoanFee - gasCost - swapSlippage
If profit < threshold (usually $50–100 considering risk), skip.
How We Build an Efficient Bot?
- Analyze Compound v3 architecture and deploy a test environment (Goerli/Sepolia).
- Develop event-based monitoring with Redis and health factor calculator.
- Implement liquidator smart contract with flash loan.
- Comprehensive testing on a mainnet fork.
- Deploy and configure monitoring (alerts, logs).
- Documentation and source code transfer.
Atomicity of Calls Is Critical
Separate absorb and buyCollateral calls create a window for frontrunning. Another MEV bot could notice the Absorb event and intercept the collateral purchase. Atomicity in a single transaction (via an aggregator contract) eliminates this risk.
Common Mistakes and Their Solutions
| Mistake |
Solution |
| Ignoring cooldown after absorb |
Retry logic with backoff |
| Incorrect minimum output calculation |
Dynamic calculation based on oracle price with 1–2% buffer |
| Not checking protocol reserves |
Check getReserves() before sending |
| Lack of atomicity |
Use an aggregator contract for a single call |
What Is Included in Development?
- Full audit of the liquidation model and writing of the liquidator smart contract (Solidity, with Foundry tests).
- Monitoring service in TypeScript or Rust with event-based subscription and Redis.
- Integration with Flashbots for private mempool.
- Testing on a mainnet fork and dry-run on mainnet.
- Deployment and configuration of alerts (Telegram, Slack).
- Documentation and source code transfer with explanations.
- Support for one month after launch.
Timeline and Cost Estimates
A basic bot for Compound v3 on a single chain takes 2.5–3 weeks. A multi-protocol bot (Compound + Aave + Euler) with MEV optimization — 6–8 weeks. Implementation for Arbitrum/Base adds 1 week per chain. Cost is calculated individually; we provide a project estimate within 2 days.
Contact us for a preliminary estimate of your project — gain a competitive advantage in the liquidation market. We guarantee quality and adherence to deadlines. Get a consultation on configuring an MEV bot.
DeFi Protocol Development
We design modular DeFi protocols where the math of stablecoins, liquidity, and oracles works flawlessly. Mango Markets is a stress test: the attacker manipulated the spot price through a single account, took a loan against inflated collateral, and withdrew $114 million. The oracle took the price from a single source without TWAP. Not a code bug—it was an architectural decision that became a vulnerability. Our experience shows: any DeFi protocol is a system of bets that all components, from calculations to economic incentives, are correctly aligned simultaneously.
We don't write code under the 'if it works, don't touch it' mindset. We model stress scenarios: cascading liquidations, depegs, flash loans. Only then do we build events that won't break the protocol.
Why are oracles a critical component of DeFi?
Most major DeFi hacks started with oracle manipulation. Let's break down the three layers we use in every project.
Spot price as oracle—not an option. Uniswap v2 spot price can be shifted by a flash loan in one transaction. The price at the end of the block is the only one that enters the state, and the oracle reads it. Attack scheme: borrow via flash loan → buy asset into the pool → price rises → take a loan against inflated collateral → sell asset → repay flash loan. One transaction.
TWAP as protection. Uniswap v3 observe() averages the price over a period (30 minutes). Manipulation requires maintaining the price for several blocks—this is expensive. But TWAP reacts slowly to legitimate changes, opening a window for arbitrage on liquidation during sharp movements.
Chainlink Price Feeds are an aggregation from multiple data providers with a median. Standard for lending. Problem: heartbeat 1–24 hours and deviation threshold 0.5%. If the price doesn't move, the feed may not update for a day. In volatile markets—lag.
| Oracle |
Mechanism |
Manipulation Protection |
Latency |
| Chainlink |
Median from independent providers |
High (decentralization) |
Up to 24h at 0% movement |
| Uniswap v3 TWAP |
Average price over N blocks |
High (hard to maintain) |
30 min – 1 h |
| Pyth Network |
Cross-chain low-latency |
Medium (dependent on publisher) |
Seconds |
In production, we use a two-tier check: Chainlink aggregator + Uniswap v3 TWAP as a verifier. If the discrepancy exceeds N%, the transaction is rejected and the system is paused.
How to protect a DeFi protocol from flash loan attacks?
Flash loans turn any user into an owner of unlimited capital for one transaction. Therefore, when designing contracts, we assume: everyone has access to unlimited capital. This completely changes the threat model.
Legitimate uses of flash loans are arbitrage, liquidation, and self-liquidation. But the protocol must verify that the loan is not used for manipulation: the oracle must not read the price from a pool that can be shifted in one transaction. We add checks on block.timestamp and minimum liquidity depth.
Key Components of DeFi Architecture
| Protocol Type |
Core Mechanism |
Main Risk |
| DEX (AMM) |
x*y=k or concentrated liquidity |
impermanent loss, oracle manipulation |
| Lending |
collateral ratio, liquidation |
bad debt during cascading liquidations |
| Yield aggregator |
auto-compounding strategies |
rug via strategy upgrade |
| Derivatives / Perps |
funding rate, mark price |
liquidation cascades, socialized losses |
| Liquid staking |
stETH-style rebasing |
depegging on mass unstake |
AMM: From x*y=k to Concentrated Liquidity
Uniswap v2 uses x * y = k. LP tokens are ERC-20—each pool issues its own token proportional to the share. Problem: liquidity is spread across the entire curve, most of it unused.
Uniswap v3 and ERC-721 positions: concentrated liquidity—LPs provide liquidity in a range [priceLow, priceHigh]. Capital efficiency up to 4000x for stable pairs. But ERC-721 breaks vault strategies built for ERC-20. Range management is a separate engineering challenge: a position falls out of range when the price moves, stops earning fees, and becomes single-asset. Protocols like Arrakis Finance automatically rebalance. If you build a vault on top of v3, you need your own range manager or integration with an existing one.
Slippage in v3 is calculated via sqrtPriceX96—96-bit fixed-point math. Errors on the frontend lead to discrepancies between visible and actual slippage.
Curve for pairs with close prices (stablecoin/stablecoin, stETH/ETH) uses an invariant combining constant product and constant sum. Lower slippage within the peg range. Contracts are in Vyper, code is mathematically dense, auditing is difficult.
Lending Protocols: Collateral, Liquidation, Bad Debt
LTV defines the maximum loan against collateral. Liquidation threshold is the level for liquidation. The difference is the buffer for the liquidator. Typical example: LTV 75%, liquidation threshold 80%, bonus 5%. If the price drops 20%+, the position is open for liquidation.
Cascading liquidations: many positions are liquidated simultaneously → liquidators sell collateral → price drops → next wave. LUNA/UST 2022 is a classic cascade.
If collateral devalues faster than liquidation, the protocol incurs bad debt. Aave uses a Safety Module (staked AAVE), Compound uses reserves. Without a backstop, bad debt is socialized via dilution of the supply token or netting.
Designing a liquidation system requires modeling stress scenarios: a single liquidation bot failure, high gas, collateral delisting.
Yield Farming and Incentive Mechanics
Liquidity mining distributes governance tokens to LP providers. Problem: mercenary capital—farmers come, sell tokens, leave. TVL is illusory.
Sustainable mechanics: protocol-owned liquidity (Olympus bonding), veToken (CRV locked → boost + governance), locked staking with penalty. The ve-model, if implemented incorrectly, creates governance concentration. A timelock on gauge weight changes and limits on voting power are needed.
What Our DeFi Protocol Development Includes
- Architectural documentation: contract interaction diagrams, liquidation stress tests, oracle calculations.
- Implementation in Solidity 0.8.x with OpenZeppelin 5.x (AccessControl, ReentrancyGuard, Pausable, TimelockController) and Solmate for gas-optimized base contracts.
- Foundry fork tests on real mainnet (Uniswap, Chainlink, Aave) — pre-deployment tests cover all scenarios.
- Audit: at least two independent auditors for TVL over $1M. Code4rena or Sherlock for bug bounty.
- Deployment with Gnosis Safe 3/5 multisig + timelock 48–72 hours.
- Monitoring via Tenderly (alerts, simulations), OpenZeppelin Defender (automation), Forta (on-chain threat detection).
- Post-launch support: updates, patches, upgrades via proxy.
Our Expertise and Experience
We have been developing DeFi protocols since 2020, delivering 30+ projects with a combined TVL of over $150 million. Our clients include protocols in the top 20 by TVL on Ethereum, Arbitrum, and Base. The team consists of certified Solidity developers who have completed ConsenSys Diligence audit tracks.
DeFi basic principles that we apply in practice.
Timelines
- DEX with AMM (Uniswap v2 fork): 6–10 weeks
- Lending protocol (Aave-style, single collateral): 3–5 months
- Yield aggregator with multiple strategies: 2–4 months
- Full-fledged DeFi protocol with governance: 5–8 months including audit
Cost is calculated individually—contact us for a project estimate.
Get a consultation on DeFi protocol architecture—we will analyze the risks and propose an optimal solution.
