Most of the coverage treated these as contagion events, which they were. But contagion describes how losses spread, not why some positions were exposed, and others weren’t. That question has a more useful answer, and it starts with how the industry talks about yield: as a product, something you market, package and sell like a phone plan, then stack against your competitor’s number and let users sort by size. That framing makes yield easy to compare. It also strips out almost everything that determines whether a given yield holds up.
On any exchange interface, yield arrives as a figure on a list: 8%, 24%, whatever the next protocol needs to advertise. The figure is usually accurate and close to uninformative. Behind it sits a system: where capital is deployed, how risk is priced, what happens at redemption, who holds custody, which regulatory perimeter applies and what the underlying assets are. The annual percentage yield is that system’s output. Read that alone, and it tells you the system was functioning at the moment the number was generated, and nothing about how it behaves under stress.
Kelp’s rsETH shows how far a headline figure can drift from what’s underneath it. It was sold as a yield-bearing liquid staking token. Still, structurally it was a token with 20 bridge dependencies and a single-verifier configuration that needed multiple human operators to remain uncompromised, backed by assumptions most of its users had no way to read. The oracle kept valuing it at par long enough for the attacker to borrow $190 million against fabricated supply before Aave could freeze the market. The failure was architectural, several layers below the number holders were pricing off.
