Curve

In July 2023 about $73.5m was drained from four Curve pools. The bug was a reentrancy flaw in specific versions of the Vyper compiler — the tool that turns source code into deployed bytecode — and not in anything Curve wrote.

That distinction matters and it does not absolve. A protocol is responsible for its dependencies: choosing a language, pinning a compiler version and reviewing what that toolchain produces are all part of shipping contracts that hold other people’s money. The failure was real. It was a supply-chain failure rather than a design failure, which is the same category as the compromised signing tool that cost Bybit $1.4–1.5bn.

Roughly 73% of the funds came back within a week, because two of the attackers negotiated and returned them. That is a genuinely good recovery rate and it rested on the goodwill of people who had just stolen the money, which is not a control anyone should count on twice.

We score its record 5 out of 10 — middling, reflecting a real loss with a substantial recovery, against protocols like Aave that have had no depositor-loss event at all.

Curve’s importance is not captured by its own volume. Its pools are where stablecoins and pegged assets actually trade against each other, which means the depth sitting in them is what holds several other tokens near their peg — including assets we rate elsewhere on this site.

That makes it a systemic dependency in a way a comparably sized general-purpose exchange is not. If Curve depth thins, the effect shows up in the peg stability of things that are not Curve, and the people holding those assets will mostly not know why.

Coverage scores 9 out of 10 on the DeFi card, reflecting how widely deployed it is. Capital efficiency scores 3 out of 10 on the DEX card, the weakest criterion it carries: the stableswap design that makes pegged assets trade tightly does so by holding a great deal of liquidity relative to the volume it turns over. That is a trade, not a flaw — the depth is the product.