When the principal becomes a risk variable in DeFi lending, how does the algorithm reshape the market?

Author: @0xJaehaerys, Gelora Research

Original compilation: EeeVee, SpecialistXBT, BlockBeats

Original title: Lending Market Doesn’t Need a Curator Model

Editor’s note: After Stream Finance and USDX experienced collapses, the DeFi community is undergoing a painful disillusionment. Protocols like Morpho and Euler introduced the “Curator” model, originally intended to solve liquidity fragmentation, but inadvertently brought human moral hazard back onto the chain. The author points out that current lending protocols wrongly conflate “risk definition” with “order matching.” By drawing on traditional finance’s order book model, this article constructs a new paradigm of algorithmic routing that requires no curator.

Evolutionary Logic of the Lending Market

Looking back at the evolution of on-chain trading can provide references for understanding the lending market.

AMMs based on constant functions (like Uniswap) solve a fundamental problem: how to create a market without active market makers. The answer is to preset the “shape” of liquidity using an invariant function. Liquidity providers agree in advance on a strategy, and the protocol automates execution.

This works well in trading because transactions are relatively simple: buyers and sellers meet at a certain price. But lending is much more complex. A single loan involves multiple dimensions:

Interest rate

Collateral type

Loan-to-Value ratio (LTV)

Term (fixed vs. variable)

Liquidation mechanism

Matching loans requires satisfying all of these constraints simultaneously.

Early DeFi lending directly adopted solutions similar to AMMs. Protocols like Compound and Aave preset interest rate curves, where lenders deposit into shared pools. This allows the lending market to operate without active lenders.

However, this analogy has a fatal flaw. In DEX trading, the shape of the constant function curve affects execution quality (slippage, depth); in lending, the shape of the interest rate curve directly determines risk. When all lenders share one pool, they also share all risks associated with the collateral accepted by that pool. Lenders cannot express a preference for taking on only specific risks.

In trading, order books resolve this: they allow market makers to define their own “curves.” Each market maker quotes at their comfortable prices, and the order book aggregates these quotes into a unified market. Yet each market maker still controls their own risk exposure.

Can lending adopt the same approach? A project called Avon attempts to answer this.

The Liquidity Fragmentation Challenge

To give lenders control, the first DeFi attempt was market isolation.

Protocols like Morpho Blue and Euler allow anyone to create lending markets with specific parameters: designated collateral, borrowed assets, fixed LTV, and interest rate curves. Lenders deposit into markets aligned with their risk preferences. A bad debt in one market will never affect others.

This is perfect for lenders—they get the risk isolation they want.

But for borrowers, this creates fragmentation.

Take ETH-USDC lending as an example, there could be dozens of markets:

Market B: $3 million liquidity, 86% LTV, 5.1% interest

Market C: $2 million liquidity, 91% LTV, 6.8% interest

…and nine more with lower liquidity

A user wanting to borrow $8 million can’t get it from a single market. They must manually compare prices, execute multiple transactions, manage scattered positions, and track different liquidation thresholds. Theoretically, optimal solutions involve splitting the loan into more than four markets.

In practice, no one does this. Borrowers usually pick just one market. Funds are underutilized across fragmented pools.

Market risk isolation solves the lender’s problem but creates issues for borrowers.

Limitations of the Curator Vault Model

The curator vault model attempts to bridge this gap.

Its idea is: professional curators manage fund flows. Lenders deposit into the vault, and the curator allocates funds across underlying markets, optimizing yield and managing risk. Borrowers still face fragmented markets, but at least lenders don’t have to manually rebalance positions.

This benefits lenders who prefer to “sit back,” but it introduces a form of discretion that DeFi aimed to eliminate: human decision-making.

Curators decide which markets receive funds and can reallocate at any time. Lenders’ risk exposure changes with curator decisions and cannot be predicted or controlled. As one Twitter user said: “The curator is PvP with the borrower, but the borrower doesn’t even know they are being ‘harvested.’”

This asymmetry manifests not only in strategy but also in interface accuracy. Morpho’s UI sometimes shows “$3 million available liquidity,” but in reality, low-interest funds are scarce, and most liquidity resides in high-interest zones.

When liquidity coordination relies on human decisions, transparency suffers.

Fund allocators adjust market liquidity on their schedule, not based on real-time demand. Vaults try to solve fragmentation through rebalancing, but rebalancing incurs gas costs, depends on curator willingness, and is often lagging. Borrowers still face suboptimal rates.

Separating Risk from Matching

Lending protocols conflate two distinct modules.

User-defined risk: different lenders perceive collateral quality, leverage differently.

Matching mechanism: mechanically assigning loans. It doesn’t need subjective judgment—only efficient routing.

Pool models bundle these, stripping lenders of control.

Isolation pools separate risk definition but abandon matching, requiring borrowers to manually find the best path.

The curator vault model reintroduces matching via the curator role but relies on trust in the curator.

Is it possible to automate matching without introducing discretion (manual intervention)?

Order books in trading have achieved this. Market makers define quotes, the order book aggregates depth, and matching is deterministic (price priority). No one decides where orders go; the mechanism determines everything.

Applying this principle to lending: CLOB (Central Limit Order Book) lending:

Lenders define risk via isolated strategies.

Strategies publish quotes to a shared order book.

Borrowers interact with a unified liquidity pool.

Matching happens automatically, without curator intervention.

Risk remains with lenders, coordination becomes mechanical. No third-party trust is needed at any step.

Two-tier Architecture

Avon implements order book lending through two distinct layers.

Strategy Layer

A “strategy” is an isolated lending market with fixed parameters.

Strategy creator defines: collateral/asset pair, liquidation LTV, interest rate curve, oracle, liquidation mechanism.

Once deployed, the interest rate curve shape cannot be changed. Lenders know the rules upfront.

Funds never move between strategies.

Depositing into Strategy A means your funds stay there until withdrawal. No curator, no rebalancing, no sudden risk exposure shifts.

While parameter setting is handled by a “strategy manager,” they are fundamentally different from curators: curators are fund allocators (decide where the money goes), strategy managers are risk controllers (define rules but don’t move funds), similar to Aave DAO. The decision-making power remains with lenders.

How does the system adapt to market changes? Through competition, not parameter adjustments. If the risk-free rate spikes, old strategies are phased out (funds flow out), and new strategies are created (funds flow in). “Discretion” shifts from “where the funds go” (curator decision) to “which strategy to choose” (lender decision).

Matching Layer

Strategies don’t directly serve borrowers. Instead, they publish quotes to the shared order book.

The order book aggregates all strategies’ quotes into a single view. Borrowers see combined depth across all strategies accepting their collateral.

When a borrower places an order, the matching engine:

Filters quotes for compatibility (collateral type, LTV requirements).

Sorts by interest rate.

Matches starting from the lowest rate.

Settles atomically.

If one strategy can fulfill the entire order, it does so; otherwise, the order is automatically split across multiple strategies. Borrowers perceive only one transaction.

Important: The order book only reads strategy states; it cannot modify them. It only coordinates access, with no authority to allocate capital.

The Gospel of RWA

DeFi has long faced a structural dilemma in institutional adoption: compliance requires isolation, but isolation kills liquidity.

Aave Arc experimented with “walled garden” pools—compliant participants have their own pools. Result: shallow liquidity, high spreads. Aave Horizon explored “semi-open” models (RWA issuers KYC, lending permissionless)—progress, but institutional borrowers can’t access Aave’s main pool of $32 billion. Some projects are exploring permissioned rollups. KYC done at infrastructure layer. This suits some use cases but disperses liquidity across networks. Users compliant on chain A can’t access liquidity on chain B.

Order book models offer a third way.

At the strategy layer, access control (KYC, regional restrictions, accredited investor checks) can be implemented arbitrarily. The matching engine only handles pairing.

If a compliant strategy and a permissionless one offer compatible terms, they can fill the same loan simultaneously.

Imagine a corporate treasury tokenizing government bonds to borrow $100 million:

$30 million from a KYC-verified institutional pool (pension fund LP)

$20 million from an accredited investor pool (family office LP)

$50 million from a permissionless pool (retail LP)

Funds never physically mix at source; institutions stay compliant, but liquidity is globally unified. This breaks the deadlock of “compliance equals isolation.”

Multi-Dimensional Matching Mechanism

Order books only match on one dimension: price. Highest bid matches lowest ask.

Lending order books must match on multiple dimensions:

Interest rate: must be below the borrower’s maximum acceptable rate.

LTV: borrower’s collateralization must meet strategy requirements.

Asset compatibility: currency match.

Liquidity: sufficient market liquidity.

Borrowers providing more collateral (lower LTV) or accepting higher rates can match with more strategies. The engine finds the cheapest path within these constraints.

For large borrowers, one caveat: in Aave, $1 billion liquidity is a single pool. In order book lending, $1 billion can be spread across hundreds of strategies. A $100 million loan can quickly deplete the entire order book, starting from the cheapest strategies, gradually filling toward the most expensive. Slippage is obvious.

Pool-based systems also have slippage, but in different form: surge in utilization pushes up interest rates. The difference is transparency. In order books, slippage is known upfront. In pools, it only manifests after execution.

Floating Rates and Requoting

DeFi lending uses floating interest rates. As utilization changes, rates adjust accordingly.

This creates synchronization challenges: if strategy utilization shifts but quotes on the order book aren’t updated, borrowers may transact at outdated prices.

Solution: continuous requoting.

Whenever a strategy state changes, immediately post new quotes to the order book. This requires infrastructure with:

Fast block times.

Low transaction costs.

Atomic state reads.

That’s why Avon is built on MegaETH. On mainnet Ethereum, this architecture is infeasible due to high gas costs.

Friction points:

If market rates change but fixed curves don’t adapt, a “Dead Zone” occurs—borrowers find the rates too high and don’t borrow; lenders earn nothing. In Aave, curves auto-adjust. In CLOB mode, lenders must manually withdraw and migrate to new strategies—costly for control.

Multi-Strategy Position Management

When a loan is funded via multiple strategies, the borrower effectively holds a multi-strategy position.

Although it appears as a single loan, underlying components are independent:

Independent interest rates: component A’s rate may rise due to utilization, component B remains unchanged.

Independent health ratios: if asset prices fall, components with stricter LTV are liquidated first. Instead of a single liquidation, there’s a series of partial liquidations, akin to “being eaten away.”

To simplify user experience, Avon offers unified position management (one-click collateral addition, automatic proportional allocation) and one-click refinancing (using flash loans to borrow and repay, always locking in optimal rates).

Conclusion

DeFi lending has evolved through several phases:

Pool-based protocols (Pooled): offered deep liquidity to borrowers but removed control from lenders.

Isolated markets (Isolated): gave control to lenders but fragmented experience.

Vaults (Vaults): attempted to bridge both but introduced human decision risks.

Order book lending (CLOB): decouples the previous models. Risk definition reverts to lenders, with matching powered by order book engines.

The core principle is clear: when matching can be fully implemented in code, human intervention becomes unnecessary. Markets can self-regulate.