There is a number buried in LlamaRisk's October 9 snapshot that should stop anyone who has ever written a liquidation bot mid-sentence. On Aave's Arc deployment, the largest syrupUSDC positions are carrying health factors between 1.01 and 1.03. If you do not live inside these dashboards, here is the translation: a health factor of 1.00 is the exact line where a position flips from safe to seizable. At 1.03, a borrower is roughly three percent of adverse price movement away from being handed to the liquidators.
On its own, that is not a scandal. E-mode markets are built to run hot, and that is the entire purpose of efficiency mode: to let tightly correlated assets borrow close to the full value of their collateral. What made me put my coffee down was not the health factor at all. It was the exit. The collateral sitting behind those positions cannot be sold in a single transaction. It has to be bridged off Arc, then redeemed through a queue on Ethereum, and the local market that could absorb an immediate sale holds something like five hundred thousand dollars of depth.
Five hundred thousand. Against positions the same snapshot values in the tens of millions.
I have audited enough of these structures to know that in DeFi, danger rarely announces itself as a bug. It arrives as a mismatch that nobody wrote down. This is one of those mismatches, and it is worth understanding in full, because the pattern it exposes is going to repeat across every new chain Aave touches.
Context: how a lending market ends up holding collateral it cannot sell
To understand why this matters, you have to understand what Aave has quietly become. The protocol stopped being a simple deposit-ETH-borrow-stablecoin machine years ago. It is now a collateral aggregator, a hub that accepts an ever-widening universe of yield-bearing assets and lets people borrow against them. On Monad and Arc, the newest frontiers, that universe now includes Pendle's principal tokens, PT-AUSD among them, and Maple's syrupUSDC vault shares.
This is a genuinely impressive act of composability. Pendle splits a yield-bearing asset into a principal token and a yield token, and the principal token trades at a discount to its eventual redemption value, converging to par at maturity. Maple's syrupUSDC is a share in an institutional credit vault that earns yield from real-world lending. Both are legitimate financial instruments. Both, in the right hands, are exactly the kind of asset a sophisticated borrower wants to post as collateral.
But here is the thing the marketing decks glide past. When Aave accepts these assets, it is not just importing a price. It is importing an exit. And the exit for a Pendle PT or a Maple vault share is nothing like the exit for ETH or USDC.
A Pendle PT does not have a native redemption path until maturity. Before expiry, if you want your money out, you sell it into a Pendle AMM pool, which in the current configuration of these markets is composed of roughly 47% PT and 53% SY, the standardized yield-bearing wrapper. After maturity, you can redeem directly through the SY. A Maple vault share is different again: there is no native redemption on Arc at all. To get your money out you must bridge the share back to Ethereum, a process that takes two to five minutes at a throughput of roughly ten million dollars an hour, and then enter a first-in-first-out redemption queue, where most requests clear within twenty-four hours but the tail can stretch to thirty days.
Read those two sentences again, because they are the heart of everything that follows. The asset Aave uses to value your collateral, and the asset a liquidator uses to recover value, do not leave through the same door. And when the doors diverge, a health factor stops being a safety margin and becomes something closer to a rumor.
I learned this lesson in the least glamorous way possible. In 2020, working as a junior community analyst at Aave during DeFi Summer, I spent weeks explaining EIP-1559 to people who were panicking about fee burning. What I took away was not a technical insight but a human one: when people cannot see the exit, they invent one in their heads, and the invented exit is usually worse than the real one. The same is true of collateral. If the exit is slow, the market prices that slowness somewhere. Often, it prices it as a surprise.
Core: the oracle says one thing, the market says another
Here is the mechanical root of the problem, and it is subtle enough that it took me a full pass through the LlamaRisk data to see it clearly.
Aave values collateral through oracles. For a Pendle PT, the oracle in question is a linear discount oracle. This is a clever design. Instead of reading the PT's spot price from an AMM, the oracle projects the PT's price along a predictable path that converges toward par as maturity approaches. The line is smooth, it is independent of the AMM's spot price, and under normal conditions it is resistant to manipulation. You cannot shove the oracle around by dumping into a thin pool.
That resistance is exactly what makes it dangerous in a stress event.
Think about what happens when real selling pressure arrives. The AMM spot price for the PT drops, because that is what selling pressure does. But the linear discount oracle does not follow it down, because the oracle is not watching the AMM. It is watching the calendar. So now you have two prices that have quietly divorced: the oracle price, which is what Aave uses to decide whether your position is safe and how much a liquidator is allowed to seize, and the AMM price, which is what a liquidator actually receives when they try to convert that seized collateral into cash.
A liquidator facing that gap is being asked to buy at the oracle's number and sell at the AMM's number. The difference is a guaranteed loss. A rational liquidator does not take a guaranteed loss. A rational liquidator walks away.
And when liquidators walk away, positions that should have been closed simply stay open, accruing interest, drifting further underwater, and quietly converting a solvency question into a bad-debt question.
This is not a code vulnerability. There is no reentrancy bug here, no integer overflow. It is a design mismatch: Aave prices collateral with an oracle, but liquidators recover value through an exit mechanism, and the two are not required to agree. When they disagree, the health factor is nominal, not real. It describes a safety margin that exists on paper and evaporates on contact.
I have spent years translating dense cryptographic logic into plain language, starting with a tool I built in 2017 that turned whitepapers into summaries for non-technical students, and this is one of those cases where the plain language is more damning than the technical one. In ordinary speech: Aave is telling you your house is worth a million dollars, while the only buyer in town is offering half a million and will not close for three weeks. The appraisal is not wrong about the neighborhood. It is just answering a different question than the one that matters when you have to sell.
The E-mode trap: why 1.03 is not the anomaly
The second piece of the puzzle is why these positions sit so close to the line in the first place. The instinctive read is that the borrowers are reckless. The more accurate read is that the borrowers are doing exactly what the parameters invite.
Aave's efficiency mode lets correlated assets borrow against each other with far less capital cushion than a normal market requires. On these stablecoin-correlated positions, the loan-to-value cap sits at 93% and the liquidation threshold at 95%. Do the arithmetic. If you borrow to the cap, which is precisely what the parameters are designed to let you do, your health factor settles naturally in the neighborhood of 1.02. A fully leveraged position in this configuration cannot have a comfortable health factor. The math forbids it.
That is why the snapshot's median health factor of 1.03 should not be read as a few risky borrowers. It is better read as the market operating as designed. The health factor is low because the design makes it low. The real question was never whether the health factor is small. The real question is whether the exit mechanism can support a health factor that small, and on Arc and Monad, it cannot.
I want to be fair to the design here, because I have spent enough time in risk committees to know these parameters are not chosen carelessly. E-mode assumes correlated assets move together, so a 95% liquidation threshold is not reckless when the collateral and the debt are two flavors of the same stablecoin. The assumption is reasonable. The problem is that it is an assumption about price, and it silently smuggles in an assumption about liquidity, that a position this leveraged can always be unwound. Those are not the same assumption. In a market where the exit takes hours to days, they are not even close.
The liquidation bonus that does not cover the cost of liquidating
Now bring in the incentive. The liquidation bonus in these stablecoin E-mode markets is 2.62%. That is the liquidator's pay: seize 2.62% more collateral value than the debt they repay.
Against a normal collateral asset, something you can sell in a single swap on a deep pool, 2.62% is thin but workable. Against this collateral, it is a rounding error fighting an avalanche of costs. Let me itemize what a liquidator actually pays to close one of these positions.
Bridge time. Getting syrupUSDC off Arc takes two to five minutes, at roughly ten million dollars an hour of throughput. That is not just latency; it is a cap on how much can be moved at once, which means a large liquidation physically cannot be exited quickly even if a liquidator wanted to.
Queue time. Once on Ethereum, the redemption enters a FIFO queue. Most requests clear within a day. The tail runs to thirty days. Thirty days of capital lockup, during which the asset's price, and the borrower's situation, can move again.
AMM slippage. The local syrupUSDC market on Arc, the one that could absorb an immediate sale, saturates at around five hundred thousand dollars on the USDC side. Sell more than that and the price walks down hard. A liquidator selling a multi-million-dollar position into a five-hundred-thousand-dollar pool is not liquidating; they are cratering their own proceeds.
Oracle-to-AMM spread. As established, the liquidator buys at the oracle's price and sells at the AMM's price. In stress, that spread is a loss, not a gain.
Stack those four costs against a 2.62% bonus and the conclusion is unavoidable. In an extreme scenario, the expected value of liquidating one of these positions can go negative. A liquidator who does the math will not show up. And a liquidation market with no liquidators is not a market at all, it is a queue of bad debt waiting to be recognized.
I have seen this movie before, in a smaller theater. In 2022, after FTX, I founded Resilience DAO to help displaced Web3 workers find their footing, and one of the recurring conversations was with liquidators and market makers who had quietly stopped bidding during the worst of the cascade. They did not stop because they had lost faith in the protocols. They stopped because the reward for taking risk had fallen below the cost of bearing it. That is the exact dynamic lurking in this 2.62% number.
The liquidity mismatch: two orders of magnitude apart
If you want a single image that captures this whole situation, hold these two numbers next to each other.
On Arc, Aave's Core Hub holds 143.45 million dollars of USDC. Of that, 83.82 million has already been withdrawn, leaving 59.63 million available. That is a deep, healthy-looking lending market. Depositors can borrow. The books look strong.
Now look at the other side. The local market where a liquidator would have to sell seized syrupUSDC has roughly five hundred thousand dollars of depth.
Fifty-nine million dollars of borrowable liquidity. Half a million dollars of absorbable collateral. Those numbers are roughly two orders of magnitude apart.
This is the mismatch that the health factor cannot see. Aave measures the safety of a position against the price of its collateral, but the borrower's debt is repaid, and the protocol made whole, only when that collateral is actually sold. The lending side has liquidity. The liquidation side does not. And no parameter in the risk framework captures that gap, because the framework was built to reason about prices, not exits.
It is worth pausing on how general this pattern is, because Aave is not unique here. Every lending market that expands its collateral set faster than it deepens its liquidation markets inherits the same flaw. Morpho's isolated markets, Compound's isolated pools, Aave's spokes: the architecture differs, the failure mode rhymes. The industry has gotten very good at measuring how much can be borrowed and much less good at measuring how much can be sold. When I look at the next wave of lending deployments, I am no longer asking what the TVL is. I am asking who the buyer is when the music stops, and how many dollars deep that buyer is willing to go.
Concentration: two positions, ninety-seven percent
There is a further wrinkle that turns a structural problem into a fragile one. On Arc, two positions account for roughly 97% of the supplied syrupUSDC.
Let that land. The entire syrupUSDC collateral base on this deployment is effectively two wallets. That cuts in two directions, and both are worth holding in mind.
The alarming direction is mechanical. If those two positions ever need to be liquidated, they do not need to overwhelm the five-hundred-thousand-dollar pool on their own. They only need to be large relative to it, which they are, by an enormous multiple. The first dollars of selling would crater the price, which would push the health factor down further, which would trigger more liquidation, which would crater the price again. That is a cascade, and the local pool has no capacity to interrupt it.
The reassuring direction is behavioral. Two wallets that large are not anonymous retail. They are almost certainly a single sophisticated entity or a very small group, and a single sophisticated entity can do something a crowd cannot: coordinate its own deleveraging. A whale who sees the cliff coming can walk away from it deliberately, unwinding in size and over time, before the forced mechanism ever engages. Concentration risk and coordination capacity are the same fact viewed from two angles. Which one dominates depends entirely on whether the whale is paying attention.
I have a professional bias toward assuming whales are paying attention, but also a professional memory of how often they are not. The 2022 cascade was full of sophisticated desks that had the tools to exit early and the confidence to wait just a little longer.
Upstream dependency: Aave is only as safe as the assets it imports
There is a deeper layer here, and it is the one that keeps me up at night as an ecosystem thinker rather than a mechanism-watcher. Aave did not build the collateral in question. It imported it. The Pendle PT is priced by Pendle's mechanisms. The Maple vault share is redeemed according to Maple's terms. Aave sits in the middle, holding assets whose exit conditions are set by other protocols.
That makes Aave a hub with upstream dependencies, and upstream dependencies transmit shocks. If Pendle's AMM pools thin out, the PT exit gets worse, and the liquidation math on Aave degrades with it. If Maple's redemption queue congests, say because its underlying credit book hits turbulence, the syrupUSDC exit slows, and again the Aave positions get harder to unwind. Aave's liquidation feasibility is not a property of Aave alone. It is a property of every protocol in the chain, and Aave does not control most of them.
This is the double edge of composability, and it is worth naming precisely. When you stack yield-bearing assets into a lending market, you are not diversifying risk, you are concentrating it. The same asset that earns yield upstream is the asset that gets liquidated downstream. If that upstream yield is itself credit risk, and in Maple's case it is, then the collateral's value and the borrower's income are two expressions of the same underlying exposure. They will fail together, not separately.
For Maple specifically, this deserves a flag. syrupUSDC's yield comes from institutional credit. If that credit book sours, two things happen at once: the vault share loses value, and the yield that made it attractive to hold disappears. Aave's collateral and Aave's borrower get hit by the same event. That is not diversification. That is correlation in a trench coat.
There is a quieter regulatory layer stacked on top of the credit layer, and it is easy to miss because nothing about this story looks like a compliance story. The debt and collateral here are stablecoin-denominated, USDC, USDT0, and AUSD among them. Those are regulated or semi-regulated instruments. If an issuer exercises a freeze, or a sanctions regime reaches one of these tokens, the collateral can become technically redeemable and practically un-sellable at the same time. I have spent a lot of 2024 explaining to bank executives that on-chain transparency and privacy are not opposites, and one of the lessons I carried back into DeFi is that compliance risk does not announce itself as a legal problem. It shows up as an exit that stops working. A bridge that has been working all year can stop clearing because a token on one side of it got frozen, and suddenly a liquidation path that looked mechanical becomes a negotiation with an issuer.
The growth-versus-prudence tension nobody wants to say out loud
Here is where the snapshot gets uncomfortable in a way that is not purely technical.
On Monad, the PT supply cap, thirty million dollars, has already been fully used. LlamaRisk's recommendation, in the same body of work that documents these near-liquidation positions, is to double that cap to sixty million. Read that pairing slowly. The risk service is simultaneously documenting positions that sit a hair from liquidation and recommending that the protocol accept more of exactly this kind of exposure.
I want to be careful not to read malice into this, because I do not think it is malice. Risk services operate inside a growth mandate; a lending protocol that refuses to grow loses to the one that does. But the tension is real, and it is the kind of tension that looks obvious in hindsight and invisible in the moment. The positions with the lowest health factors are precisely the new ones: the ones on the new chains, with the new collateral, under the new caps. Growth and risk are landing on the same address.
There is a detail in the framing that I find revealing. The documentation refers to a Core Hub, which reads to me like the vocabulary of Aave's V4 architecture, the Liquidity Hub and Spoke model that separates liquidity from the markets that consume it. If that is the context, then what we are looking at is the new architecture being stress-tested at the exact moment its liquidity is thinnest, on chains that are youngest, with collateral that is hardest to exit. That is not a reason to panic. It is a reason to be precise about what early means. Early does not only mean high upside. It also means thin, and fragile in ways the parameters do not measure.
A word on the tooling, because it matters
I would be lying if I said the composability on display here does not impress me. Arc's local Uniswap V4 pool, with its hooks, is exactly the kind of programmable Lego that makes DeFi genuinely powerful. V4 lets you attach custom logic to a pool, dynamic fees, custom oracles, limit orders, and that is a real leap in expressiveness.
But I have watched enough tooling waves to know the pattern. Every time a platform becomes dramatically more expressive, the population of developers who can safely use it shrinks. V4 hooks are powerful and unforgiving; the complexity spike will quietly filter out the majority of builders who try. The same is true of the exit machinery in this story. Linear discount oracles, PT and SY redemption paths, FIFO queues, cross-chain bridges: each is a legitimate tool, and each is a place where a subtle mistake compounds. The sophistication that makes these markets possible is also the sophistication that makes them hard to audit, hard to reason about, and hard to rescue when something goes wrong.
And on the cross-chain side, I have to be honest about something the ecosystem keeps trying to wish away. Ethereum's Dencun upgrade did lower the cost of moving value between rollups, and that part is real. But the user experience of bridging is still orders of magnitude worse than simply withdrawing from a centralized exchange. A liquidator forced to bridge off Arc and then queue on Ethereum is living inside that gap. The costs are lower than they were. The experience is still bad. And in a liquidation, bad experience is a cost that shows up on the balance sheet.
There is a related hype I keep running into, and it is worth a sentence of honesty here because it shapes how these chains are built. The data-availability layer has become the fashionable investment thesis of the cycle, and I understand why: it is legible, it is infrastructure, it sounds like plumbing. But most rollups do not generate enough data to need a dedicated DA layer at all, and building one anyway is a way of paying rent on a problem you do not yet have. Arc and Monad are not short on places to put data. They are short on buyers for distressed collateral. The industry keeps funding the pipes and forgetting the exits, and the mismatch in this snapshot is what that looks like when it reaches the lending market.
Contrarian: the scariest reading may not be the right one
Let me push back on my own alarm, because the most useful thing I can do for a reader is to hand them the strongest version of the opposite case.
It is entirely possible that the health factors of 1.01 to 1.03 are not a warning sign at all but the expected steady state of an E-mode market. I ran the numbers: at a 93% LTV and a 95% threshold, a fully leveraged correlated position mathematically must sit around 1.02. A market median of 1.03 is not a red flag; it is the design working exactly as intended. If that is the whole story, then this article is describing normal operations with an ominous soundtrack.
There is a second reason for calm. Those two positions holding 97% of the syrupUSDC are almost certainly one sophisticated entity. Sophisticated entities do not get liquidated by surprise; they manage their own leverage, and they have both the motive and the ability to deleverage gracefully before the forced mechanism engages. Concentration is a risk, but it is also a coordination advantage, and the entity holding the concentration is the entity best positioned to defuse it.
I hold both of those points sincerely. What keeps me from resting on them is that neither one addresses the thing that actually worries me. The reassuring case is about behavior, whether the whale acts. The alarming case is about structure, whether the exit exists. Those are different questions, and only one of them is under the protocol's control. Community is the only chain that cannot be broken, but community also cannot sell collateral into a five-hundred-thousand-dollar pool. Good behavior buys time. It does not buy liquidity.
And time, in a market this leveraged, is not free. Every day a position sits at a health factor of 1.03 is a day the protocol is trusting a mechanism it has not stress-tested at size. Trust like that is not a parameter. It is a bet.
Takeaway
The lesson of Arc and Monad is not that Aave is broken. It is that a health factor measures a price, and a liquidation needs an exit, and as this industry stretches into new chains and ever-more-exotic collateral, the gap between those two will keep widening until someone builds the missing piece. The next generation of risk frameworks will have to price liquidity the way they price volatility, or the liquidation cliffs will keep arriving as surprises. The builders who understand that first will be the ones whose markets survive the next stress test, and the ones who do not will learn, expensively, that a chain is only as strong as its exit.
