The evolution of onchain trading offers useful intuition for thinking about lending markets.
Constant Function AMMs solved a fundamental problem: how do you create markets without active market makers? The answer was to pre-specify the shape of liquidity using an invariant function. Liquidity providers agree to a strategy upfront, and the protocol handles execution automatically.
This worked because trading is relatively simple. Buyers and sellers meet at a price. But lending is more complex. A loan has multiple dimensions:
The interest rate
The collateral type
The loan-to-value ratio (LTV)
The duration (fixed-term vs open-ended)
The liquidation mechanism
Matching in lending requires satisfying constraints across all of these dimensions simultaneously.
Early DeFi lending adopted the AMM intuition directly. Protocols like Compound and Aave pre-specified interest rate curves, and lenders agreed to a pooled strategy. This allowed markets to function without active lenders, just as AMMs allowed trading without active market makers.
But the analogy breaks down in an important way. In trading, the shape of liquidity affects execution quality. Tighter spreads and deeper books give traders better prices. In lending, the shape of liquidity also determines risk. When all lenders share a pool, they share the risk of every collateral type the pool accepts. The lender cannot express a view on which risks they want to take.
Order books solve this in trading by letting market makers define their own liquidity shape. Each maker quotes at prices they're comfortable with. The book aggregates these quotes into a unified market, but each maker retains control over their exposure.
The first attempt to give lenders control was market isolation.
Rari Capital's Fuse,MorphoBlue, Euler, and similar protocols allow anyone to create a lending market with specific parameters: one collateral asset, one borrow asset, a fixed LLTV, a chosen oracle, a defined rate curve. Lenders deposit into markets that match their risk preferences. Bad debt in one market cannot affect another.
This works exactly as intended. Lenders got what they wanted: the ability to choose their exposure.
But borrowers got fragmentation.
Consider ETH-USDC lending. A dozen markets exist, each with different parameters:
Market A: $5M available liquidity, 80% LLTV, 4.2% rate
Market B: $3M available liquidity, 86% LLTV, 5.1% rate
Market C: $2M available liquidity, 91% LLTV, 6.8% rate
...and nine more
A borrower wanting $8M cannot fill their order from any single market. They must manually compare rates, execute multiple transactions, manage separate positions, and track different liquidation thresholds. The optimal path requires splitting across four or more markets.
In practice, this doesn't happen. Borrowers pick one market and accept suboptimal terms. Capital sits underutilized across fragmented pools.
Isolation solved the lender's problem by creating the borrower's problem.
Curated Vaults
Curated vaults attempted to bridge this gap.
The idea: professional allocators manage where liquidity flows. Lenders deposit into a vault. The curator distributes capital across underlying markets, optimizing for yield and managing risk. Borrowers still see fragmented markets, but at least lenders get exposure to multiple markets without actively rebalancing themselves.
This helped lenders who wanted simplicity. But it introduced something DeFi was designed to eliminate: discretion.
Curators decide which markets receive deposits. They can reallocate capital at any time. A lender's risk profile changes based on decisions they cannot predict or control.
"Curators PvP the borrowers, but the borrowers don't even know enough to PvP back."
The asymmetry extends beyond strategy. It shows up in basic interface accuracy:
"The morpho UI will say 'here's a $3M market and $2.96M is left to borrow at 21%.' But then also say 'the instantaneous borrow rate is 9.81%'... and then finally say 'actually, there's not 2.96M, there's only 38k supplied and it's 91.5% utilized.'"
This gap between displayed rates and actual execution reflects the fundamental issue: when coordination depends on human decisions, transparency suffers. Allocators move liquidity on their schedule, not the market's.
Vaults solved one problem (lender UX) while creating another (discretion risk). They attempted to address borrower fragmentation through rebalancing. If utilization spikes in one market, the curator should move liquidity from underutilized markets to restore equilibrium. But rebalancing costs gas, depends on curator discretion, and happens on the curator's schedule, not the market's. Borrowers still face timing gaps and suboptimal rates while waiting for reallocation that may never come.
Separating Risk from Matching
The core insight is that lending protocols conflate two distinct functions.
Risk definition answers: what terms am I willing to lend at? This is inherently subjective. Different lenders have different views on collateral quality, acceptable leverage, oracle reliability. There is no universal right answer.
Matching answers: given available supply, how do borrowers find the best terms? This is mechanical. It requires no judgment, just efficient routing.
Pooled protocols combine both. The pool defines risk parameters AND handles borrower access. Lenders lose control because these functions are bundled together.
Isolated protocols separate risk definition but abandon matching. Each market handles its own parameters, but there is no coordination layer. Borrowers must navigate manually.
Curated vaults add matching back through human allocators. But this reintroduces trust assumptions.
The question becomes: can matching be automated without discretion?
Order books provide this in trading. Market makers define their own quotes. The book aggregates quotes into unified depth. Matching is deterministic: best price wins. No one decides where orders route. The mechanism handles it.
CLOB lending applies the same principle to credit markets:
Lenders define risk through isolated strategies.
Strategies publish quotes to a shared order book.
Borrowers interact with unified liquidity.
Matching happens automatically, without curator involvement.
Risk stays with lenders. Coordination becomes mechanical. Neither requires trust in third parties.
Architecture: Two Layers
Avon implements order book lending through two distinct layers.
1. The Strategy Layer
A strategy is an isolated lending market with defined parameters.
Strategy creators define:
Collateral and borrow assets
LLTV (liquidation loan-to-value threshold)
Interest rate curve
Oracle configuration
Liquidation mechanism
Once deployed, the rate curve shape is immutable. Lenders know exactly what rules govern their capital before depositing.
Strategies are ERC-4626 compliant vaults. Lenders deposit, receive shares, earn yield based on utilization within that specific strategy.
The critical property: capital never moves between strategies.
If you deposit into Strategy A, your funds stay in Strategy A until you withdraw. There is no allocator. No rebalancing. No surprise exposure changes. Your risk is exactly what you chose.
Each strategy operates independently. A borrower can touch multiple strategies in one loan, but each component is evaluated separately. Bad debt in Strategy A cannot affect lenders in Strategy B.
Someone still decides the parameters. Strategy Managers choose rate curves, LLTVs, oracles. If they choose poorly, lenders suffer. The difference from curators: capital movement. Curators can reallocate your deposits to different markets without consent. Strategy Managers set rules once; your capital stays put.
This distinction matters. Risk curator in DeFi is a misnomer. They are actually capital allocators who decide where liquidity flows. In Avon Strategy Managers are true risk managers. They define parameters but do not move capital. The allocation decision stays with lenders.
This raises a question:if curve shapes are immutable, how does the system adapt to changing market conditions? If the risk-free rate jumps from 5% to 10%, a strategy capped at 8% becomes uncompetitive. It doesn't update. It empties. Lenders withdraw and move to strategies with better terms.
Adaptation happens through competition, not parameter changes. Inefficient strategies don't get governance votes to fix them. They lose flow. New strategies with parameters get created and capture liquidity. This means discretion shifts rather than disappears: instead of "where should capital go" (curator decision), it becomes "which strategy should I choose" (lender decision). The human element remains, just at a different point in the system.
2. The Matching Layer
Strategies do not serve borrowers directly. Instead, each strategy publishes quotes to a shared order book.
A quote represents available liquidity at a specific rate for borrowers meeting certain requirements. The strategy's rate curve determines these quotes. At low utilization, low rates. As utilization rises, rates rise.
The order book aggregates quotes from all strategies into one unified view. Borrowers see combined depth across every strategy that accepts their collateral type.
Borrowers can submit market orders (fill at best available rates), conditional orders (trigger based on parameters), or limit orders (specify maximum acceptable rate). When an order executes, the matching engine:
Filters quotes by compatibility (collateral type, LLTV requirements)
Sorts by interest rate
Fills from cheapest available
Settles atomically in one transaction
If one strategy can fill the entire order, it does. If not, the order splits across multiple strategies. The borrower receives one transaction regardless. The protocol tracks components separately underneath.
The order book only reads strategy state. It cannot modify it.
This is important. The matching layer coordinates access but has no power over capital allocation. Strategies publish. The book displays. Borrowers choose. Matching executes. No discretion anywhere in the chain.
A clarification on terminology: in trading, a filled order locks a price forever. Here, matching opens a position with a floating rate that continues to evolve. Avon acknowledges this directly: "The order book coordinates access. It never locks the rate." The term order book describes the coordination mechanism, not a guarantee of fixed terms. It's a hybrid: order book for price discovery, floating rates after execution.
Compliance Without Fragmentation
DeFi has struggled with institutional adoption for a structural reason. Compliance requires separation but separation kills liquidity.
Aave Arc tried walled gardens. Compliant participants got their own pool isolated from retail. The result was shallow liquidity, worse rates, and limited collateral options. The compliance requirement created the very inefficiency institutions were trying to avoid.
Aave Horizon improved on this with a half-open approach. RWA borrowing requires KYC from issuers but stablecoin lending is permissionless. Anyone can earn yield from institutional borrowers. This is progress but liquidity remains isolated. Institutional borrowers cannot tap into Aave's $32B+ liquidity. They are limited to whatever gets deposited into Horizon specifically.
Some projects explored permissioned rollups. These are entire chains where KYC happens at the infrastructure level. This works for some use cases but fragments liquidity at the network layer. A compliant user on Chain A cannot access liquidity on Chain B.
The order book model offers a third path.
Strategies can implement any access control logic. This includes KYC requirements, geographic restrictions, accreditation checks, or jurisdiction-specific rules. The matching engine does not care. It reads quotes, checks compatibility, and fills orders. If a compliant strategy and a permissionless strategy both offer compatible terms, both can fill the same loan.
Consider a corporate treasury borrowing against tokenized T-bills:
$30M from a strategy requiring institutional KYC (pension fund LP)
$20M from a strategy requiring accredited investor status (family office LP)
$50M from a fully permissionless strategy (retail LPs)
One transaction. One position. Three different compliance regimes. Capital stays segregated at the source but aggregates at the point of execution.
This matters for RWA adoption. Tokenized assets need institutional liquidity. Institutions need compliance. Compliance historically meant isolation. Order books break this tradeoff.
The protocol remains permissionless. Strategy-level restrictions are opt-in. Retail users never touch compliant-only strategies and compliant capital never enters permissionless vaults. But borrowers see unified depth regardless of where liquidity originated.
Multi-Dimensional Matching
Trading order books match on one dimension: price. Highest bid meets lowest ask.
Lending order books must match on multiple dimensions simultaneously:
Interest rate. The borrower's maximum acceptable rate must meet or exceed the strategy's required rate.
LLTV. The borrower's collateral ratio must meet or exceed the strategy's requirement.
Asset compatibility. Collateral and borrow token types must match.
Liquidity. The strategy must have sufficient available funds.
A match occurs only when all constraints align. This creates richer dynamics than price-only matching.
A borrower offering more collateral (lower LLTV) qualifies for more strategies. A borrower accepting higher rates qualifies for more strategies. The matching engine finds the cheapest path through this constraint space.
Example:
Borrower wants: $150k USDC, 85% LLTV, maximum 6% rate
The borrower gets $150k in one transaction at a better rate than any single strategy offered. Settlement is atomic. Collateral distributed proportionally across components.
For large borrowers, a caveat. In Aave, $1B liquidity is one monolithic pool. In order book lending, $1B might be spread across hundreds of strategies. A $100M borrow "eats through the book," filling from cheapest to most expensive. Visible slippage. Pool-based systems also have this, just expressed differently: utilization spikes push rates up. The difference is transparency. In order books, slippage is visible upfront. In pools, it materializes after execution.
Variable Rates and Requoting
DeFi lending uses variable rates. The rate changes as utilization changes. This creates a synchronization challenge.
Consider the sequence:
Strategy at 30% utilization publishes quotes at 5%
Borrower takes a loan, utilization jumps to 60%
Strategy's actual rate is now 8%
But stale quotes might still show 5%
If matching executes against stale quotes, borrowers get rates that don't reflect current conditions.
The solution: continuous requoting. As strategy state changes, it republishes quotes to the order book.
This requires infrastructure that previous protocol generations couldn't assume:
Fast blocks. Quotes must update before the next borrower arrives.
Cheap transactions. Requoting has costs. Must be affordable at high frequency.
Atomic state reads. The matching engine must see consistent snapshots.
High-throughput chains make this architecture viable. On slow expensive chains, the order book would perpetually lag behind reality. Avon builds on MegaETH specifically because the design requires sub-second state propagation.
On Ethereum mainnet, this design would be prohibitively expensive. Every utilization change requires quote updates. If a strategy has thousands of participants, any action triggers repricing. The architecture only works where infrastructure supports it.
A related concern: capital efficiency. In pool-based systems, the utilization curve automatically finds equilibrium. Rates float until supply meets demand. In order book systems with fixed curves, equilibrium isn't guaranteed.
If market rates shift but a strategy's curve doesn't adapt, a "dead zone" emerges. Borrowers won't take liquidity (too expensive), but the curve can't lower rates (it's fixed). Capital sits idle. In Aave, this resolves automatically through curve adjustment. Here, it resolves through lender withdrawal and migration to better strategies.
The system assumes competition will drive efficiency: poorly priced strategies lose flow, well-priced ones gain it. This works in aggregate but creates friction for individual lenders. If you're in a strategy that becomes uncompetitive, you earn nothing until you manually withdraw and redeposit elsewhere. The protocol doesn't rebalance for you. That's the point, but it's also the cost.
Rate volatility is higher than in pools. In Aave, a massive pool smooths rate movements. Here, if a whale exits a strategy, rates for remaining participants can spike. This is true market price discovery, not smoothed synthetic rates. Whether that's a feature or a bug depends on your perspective.
Withdrawal friction is real. If utilization hits 100% in your strategy, you cannot exit until borrowers repay. Unlike pooled systems where new deposits create exit liquidity for everyone, here you're locked to your specific strategy's utilization.
Multi-Strategy Positions
When a loan fills across multiple strategies, the borrower holds a multi-strategy position.
Each component is tracked independently. Say Alice borrows across two strategies:
Borrower
Strategy
Principal
Rate
LLTV
Health
Alice
X
$100k
3.5%
70%
1.35
Alice
Y
$50k
4.8%
80%
1.05
Independent rates. Each component's rate floats based on utilization within its strategy. Strategy X getting more borrows raises Component A's rate. Strategy Y is unaffected.
Independent health. Each component is evaluated against its strategy's LLTV.
If collateral price drops:
Component A (70% LLTV): health might fall to 1.08, still safe.
Component B (80% LLTV): health might fall to 0.84, liquidatable.
Only Component B gets liquidated. Lenders in Strategy X are completely unaffected.
Unified interface. Despite internal complexity, the borrower manages one position:
Add collateral: distributed proportionally
Partial repay: unwound in deterministic order
Full close: all components settled atomically
The protocol handles multi-strategy accounting. The borrower experiences simplicity.
But the complexity is real, and it manifests in ways that matter.
A borrower with exposure across ten strategies has ten independent health factors, ten independent rates, ten potential liquidation events. Each component has a different liquidation threshold.
When the market falls, your position gets nibbled from the edges. The conservative components with stricter limits liquidate first. As price drops further, more components fall. You do not go from healthy to liquidated in one event. You experience a sequence of partial liquidations.
In Aave you are either healthy or liquidated. Here you can end up in a degraded state that is harder to reason about.
Avon mitigates this with unified position management. You can add or remove collateral across all components of a market (e.g., all your ETH-USDC positions) in one action. The protocol uses weighted average distribution, so you're not manually adjusting each component. The complexity exists, but the interface abstracts most of it.
Refinancing has friction too. Rates float within each strategy based on utilization, so if your strategy becomes cheaper, you benefit automatically. But if a better strategy appears elsewhere, moving requires closing old positions and opening new ones. In Aave you passively benefit from any rate drop across the pool.
Avon addresses this with a planned one-click refinancing feature: the protocol repays your existing position with a flash loan and reopens it at the best available quote on the book. You get optimal rates without manual unwinding of multiple positions.
Who Is This For
Order book lending serves different users differently.
For borrowers, it's simpler than isolated markets. No manual searching across pools. The order book finds the best available terms automatically. One transaction, best execution. The complexity is abstracted away.
For lenders, there are two paths:
Active lenders (DAOs, funds, pros) choose strategies directly. Full control over risk parameters. This requires evaluation and monitoring.
Passive lenders use MegaVault. Deposit USDm, receive USDmY, earn yield automatically. The vault handles allocation.
The pro tool label applies to direct strategy selection, not the protocol as a whole. Borrowers get simpler UX than fragmented markets. Passive lenders get vault convenience. Active lenders get granular control.
Unlike Uniswap V3 where passive LPs get destroyed, here passive lenders have a dedicated path that doesn't compete with sophisticated actors.
Comparison
Property
Pool (Aave)
Isolated (Morpho Blue)
Vault (MetaMorpho)
Order Book (Avon)
Lender control
None
Full
Trust curator
Full
Borrower UX
Unified
Fragmented
Fragmented
Unified entry, complex ongoing
Liquidity depth
Real aggregation
Fragmented
Aggregated
Virtual aggregation
Risk isolation
None
Full
Partial
Full
Discretion
Governance
None
Curator
Shifts to lender (strategy choice)
Capital movement
Automatic
Manual
Curator decides
Never without consent
Large borrow execution
Rate spike via curve
Manual splitting
Depends on allocation
Visible slippage through book
Position complexity
Simple (binary outcome)
Simple per market
Simple per vault
Complex (partial liquidations)
Capital efficiency
Auto-equilibrium
Manual rebalancing
Curator managed
Competition-driven (idle risk)
Infrastructure needs
Low
Low
Medium
High (fast chains)
Target user
Passive retail
Active lenders
Passive via curator
Both (strategies or vault)
RWA/Compliance
Isolated pools (Arc, Horizon)
Per-market restrictions
Curator-managed exposure
Strategy-level, aggregated execution
Order book lending attempts to combine:
Lender control (from isolated markets)
Borrower coordination (from pooled markets)
Reduced discretion (compared to vaults)
Conclusion
DeFi lending evolved through stages:
Pooled protocols gave borrowers depth but stripped lenders of control.
Isolated markets gave lenders control but fragmented borrower experience.
Curated vaults attempted to bridge both but introduced discretion.
Order book lending separates what pools conflate: risk definition stays with lenders, matching becomes mechanical. Strategies define parameters. The order book coordinates access. Neither requires trust.
This design needs infrastructure that didn't exist when lending protocols were first built. High-throughput chains make requoting practical. Sophisticated engines handle multi-dimensional matching. Per-component tracking maintains isolation within unified positions.
The principle is clear: when matching can be expressed in code, it doesn't need humans in the loop. Markets can route themselves.
.jpg&w=64&q=75){kind=avatar}
jaehaerys
