For anyone who’s been trading or building in crypto long enough, volatility is something they’ve learned to expect. It’s pretty much part of the job description. But even with that mindset, the now infamous crash of October 10, 2025 went beyond simple “volatility” — it became a stress test that exposed what weaknesses still remain in this industry, and pushed many systems almost to the point of snapping. As a DeFi developer responsible for designing some of the systems that went through that test and endured quite well, I will say this: for all the damage it caused, the wipeout also validated some of the principles that many of us have been pushing for years. In this article, I want to take my own deeper look at why things happened the way they did, what worked, what didn’t, and what the entire ecosystem should learn from it. Setting the Stage Setting the Stage What happened on October 10 felt less like a market correction and more like a controlled demolition. Within 24 hours, Bitcoin and Ethereum both dropped more than 12% and 14% respectively, with many altcoins (AVAX, DOGE, etc.) taking even harder hits and losing upwards of 50% in value. dropped dropped Gas prices spiked above 500 Gwei, only to collapse to 0.1 Gwei two days later, mirroring the chaotic waves of market liquidity at the time. Centralized exchanges showed signs of strain. On-chain liquidity fractured as cascading liquidations raced ahead of human decision-making. 500 Gwei 500 Gwei collapse to 0.1 Gwei two days later collapse to 0.1 Gwei two days later Over $19 billion in leveraged positions were wiped out in a day. It didn’t take long for social media to start calling the whole event things like “Crypto’s Black Friday,” or “The Great Crypto Crash” — and for once, it wasn’t an exaggeration. Like I already mentioned, in crypto, you get used to volatility. Flash crashes aren’t something new. But even so, the speed and depth of this one was on a whole different level. And for the DeFi world, it wasn’t something you could rehearse for in advance. Naturally, this was also true of Curve. I’ve been saying for years that liquidation architecture determines who has better chances of surviving a crash. In this regard, October 10 provided unmistakable proof. Liquidation Models Aren’t Equal — Here’s Why Under extreme market conditions, each liquidation model is prone to breaking in its own unique way. To elaborate, let’s look at the fundamentals of the most common approaches used in crypto. CeFi: Users Have Too Little Control Centralized exchanges liquidate positions on internal oracles and proprietary logic. They are the ones to decide when your loan is underwater, how much of your collateral to sell, and at what price. Most of this happens behind closed doors, with users having very little insight or control over how any of the decisions get made. Moreover, oracles on CEXes often still rely only on data from their own orderbooks, thinking that they are the best price sources. This assumption is very wrong in the case of stablecoins, in particular, where on-chain data, in fact, often proves more accurate. The October 10 cash is clear proof of that. The entirety of it appears to have been triggered by the USDe oracle on Binance thinking that USDe depegged to as low as 60 cents. In reality, this was absolutely not the case: on Curve, for example, the USDe price stayed above 99 cents at all times. In other words, this whole event — that wiped tens of billions in liquidations and spread far beyond USDe itself — was triggered just by this faulty USDe oracle. thinking thinking stayed above 99 cents stayed above 99 cents What was the cause of the Binance oracle being faulty? It obviously used only CEX data, without taking DEX data into account. But why is CEX data so much worse? CEXs run on orderbooks, and that means that liquidity is limited in prices. There simply could be no orders beyond a certain price, and there is no floor after the asset price goes beyond this “edge” order. On AMM-based DEXs, ALL prices are supported, across the full range of possible values. So liquidity literally exists everywhere — there is no “edge” effect here, like the one in centralized orderbooks. It means that the price on platforms like Curve stays the “true” price at all times, giving a more consistent and reliable view of the market compared to centralized exchanges. Though to be absolutely fair, Binance did learn from this incident and started quickly fixing their oracles to incorporate on-chain data sources. DeFi: Transparent, But Slow Under Pressure DeFi: Transparent, But Slow Under Pressure Many DeFi protocols like MakerDAO or AAVE rely on auctions and batch liquidations. Normally, this is fine, as the approach is transparent and mathematically sound. But that model only works under normal market conditions, where certain assumptions hold true: Bots can see when a position has crossed the liquidation threshold They can bid quickly enough to win the auction The gas market stays within reasonable limits so that execution doesn’t become too expensive. Bots can see when a position has crossed the liquidation threshold They can bid quickly enough to win the auction The gas market stays within reasonable limits so that execution doesn’t become too expensive. On high-volatility days, like October 10, these assumptions easily fall through. When gas prices jump like that, auctions tend to get stuck. Overpriced bids worsen slippage, undercollateralized positions sit in auction queues until it’s too late, and users, ultimately, suffer greater losses. In practice, that’s exactly what happened, and the market was reminded of a valuable lesson — that liquidity design must account not just for “average” conditions, but for extreme ones, as well. LLAMMA: Designed Against Brittleness, But Not a Miracle Cure LLAMMA: Designed Against Brittleness, But Not a Miracle Cure Curve’s LLAMMA model was created specifically to avoid last-second full liquidation decisions. When a position starts approaching danger zones, the algorithm gradually converts collateral into its native crvUSD stablecoin. Then, if the price recovers, the conversion process can be reversed. This approach is gentler to borrowers, allowing them to avoid the proverbial “cliff” after which they stand to lose hard. And as we learned during the crash, this design held up well — not perfectly, but enough to show a notable difference between us and more typical platforms. Case Study: How LLAMMA Behaved When the Market Broke Case Study: How LLAMMA Behaved When the Market Broke The October wipeout was an extreme scenario for every DeFi protocol, and Curve was no exception in that sense. Across our LlamaLend and Mint Markets, 118 positions were liquidated, totaling $30.34 million in debt. That said, out of $235 million in total debt before the crash, $196 million remained healthy the next day — which means that over 90% of users survived. When prices took a dive, LLAMMA softened the blow by automatically moving vulnerable positions into a protective zone. Parts of collateral were converted into crvUSD, and about 22% of positions that were close to full liquidation ended up saved as a result. The result is admittedly lower than the system demonstrated in quieter periods — for example, during the August 5, 2025 downturn, we were able to save 44% of positions. But the key difference was in the state of the environment around these two events. Even well-structured systems are forced to operate at a disadvantage when the market falls as quickly as it did during the October crash. The crash also taught Curve DAO to re-adjust parameters of some markets towards more risk-averse: non-major coins as collateral should probably have much lower LTVs while Aave moved to remove high-volatility collaterals altogether. moved moved But we managed to pull through relatively well. Even as the market buckled, crvUSD maintained its peg with only minimal deviation, which was a direct result of the system holding up under maximum pressure. There was also one particular “whale” case that I feel the need to talk about. A single large wallet accounted for 72% of all liquidations as it held massive positions at a risky health factor below 5%. A ticking bomb by any measure. Still, even in this scenario, parts of the position were cushioned by our liquidation protection mechanism. But this case proves something that I feel more borrowers need to understand: no liquidation system is perfect, and it can’t save users who push risk to the extreme. Ultimately, it falls to the borrowers to protect themselves by choosing parameters wisely and in accordance with their own risk appetites. Another interesting test for LLAMMA happened in mid-November (15-16). I noticed that a large position opened back in 2023 and worth nearly $50M entered the soft-liquidation zone due to a market dip. When that happened, LLAMMA automatically converted part of his ETH into crvUSD to stabilize the loan. Later on, the market recovered, and the position exited protection intact. entered entered However, even later, the market decided to go further down, and the user’s funds were fully converted to crvUSD. The user decided to close the position at the bottom instead of waiting to be liquidated, utilizing the received funds later — probably a wise decision. This was a good example of how LLAMMA is supposed to function under more “normal” volatility conditions: stabilize the loan, buy time, and reverse when markets rebound. You don’t often see large positions behave so calmly during turbulence, but that’s the whole point of liquidation protection — the system is designed to make the volatility impact easier to bear. The Crash Exposed Industry-Wide Limits: What Still Needs Fixing? The Crash Exposed Industry-Wide Limits: What Still Needs Fixing? October 10 proved to be a DeFi-wide wake-up call. And here are some of the key lessons that I think everyone should pay attention to and internalize, regardless of which protocol they prefer. Gas volatility is a systemic risk: liquidators can’t work properly if the cost of executing a transaction jumps 10x every minute. Hopefully, new Ethereum upgrades will fix this. MEV competition pushed liquidations deeper than necessary: bots don’t care about user outcomes — they care about profitable execution conditions. CEX engines remain single points of failure: if internal oracles misprice assets, they pull the entire market with them. Gas volatility is a systemic risk: liquidators can’t work properly if the cost of executing a transaction jumps 10x every minute. Hopefully, new Ethereum upgrades will fix this. Gas volatility is a systemic risk: MEV competition pushed liquidations deeper than necessary: bots don’t care about user outcomes — they care about profitable execution conditions. MEV competition pushed liquidations deeper than necessary: CEX engines remain single points of failure: if internal oracles misprice assets, they pull the entire market with them. CEX engines remain single points of failure: Auction-based DEX systems need further design improvements: if they stall, it creates full liquidations where partial ones should’ve occurred. This crash was painful. But it validated the idea that liquidation risk should be mitigated through a structured design approach, not wishful thinking. There will certainly be more such crashes in the future, and it is important that the market learns from each one. Because these events reveal where the next improvements must be made, ultimately helping make the entire infrastructure harder to break. To end on a positive note, I’ll say this: DeFi didn’t just survive the crash — it grew stronger for it. Because now the developers have new lessons to keep in mind when raising the resilience bar across the entire ecosystem.
