Skip to main content
These are retrospective analyses, not live captures — most predate Pegana. They’re here because the shape of each failure is the shape Pegana is designed to detect.
CaseAssetClassDateWorst spreadDetected live?
USDC-SVBUSDCFiat stable2023-03-11−1300 bpsNo (pre-Pegana)
UST collapseUSTAlgo synthetic2022-05-09−7000 bpsNo (pre-Pegana)
mSOL flash crashmSOLLST2022-06-19−1500 bpsNo (pre-Pegana)
DAI / SVB contagionDAICDP2023-03-11−1100 bpsNo (pre-Pegana)
GHO mint-only floatGHOCDP2023-08-15−550 bpsNo (pre-Pegana)

​
Themes that recur

Two-numbers framing matters. USDC and DAI both had intrinsic ≈ $1 throughout the SVB weekend — only the market refused the price. mSOL had market ≈ 0.85 SOL for four minutes while Marinade’s on-chain intrinsic never moved. A monitor that watches one side prints either a false negative (intrinsic-only) or a false positive (market-only). See intrinsic vs market. Thin venues lie. mSOL’s flash crash was a single thin pool dragging an oracle. Liquidations cascaded against healthy collateral. The fix is routed quotes (Jupiter)
  • a confidence score that drops when depth disappears. See market value.
Asymmetric mechanisms fail asymmetrically. GHO floated below par for 8 months because mint was open but burn-to-par was closed. The state machine catches both short shocks and sustained drift via dwell timers. See hysteresis. CDP stables inherit collateral tail risk. DAI rode USDC down because USDC was half its reserves. Hylo’s hyUSD has SOL tail risk. The signal worth watching is the collateral ratio, not the market price. See Hylo CDP CR signal.

​
What Pegana would have shown

For every case below, we walk the timeline through Pegana’s lens — what each state transition would have looked like, what subscribers would have received, and where the alert would have been actionable vs after-the-fact. The honest answer for most of these is: Pegana would have called DRIFT 5–30 minutes before the worst print, with a follow-up DEPEG transition once the threshold crossed. That’s enough lead time to pause a liquidation engine, freeze a payout queue, or skip a swap — but it’s not magic. We can’t predict; we report.