Skip to main content
← Risk Engine Validation & Stress Testing Report

Depeg Sensitivity Evidence

Production-verified severity classification for LST and stablecoin depeg events. Supports Test 8.2. All scores on a 0.0-1.0 scale (displayed as 1-10 in the UI). For term definitions, see the Glossary. For data sources, see Data Provenance.

7-Tier Severity Classification

The stability scorer classifies depeg events into 7 tiers with geometrically doubling thresholds, half-lives, and penalty weights. Only the worst tier fires per candle — no double-counting.

TierThresholdClassificationHalf-LifeLambda
01%Minor noise1 day0.05
12%Slight deviation2 days0.10
24%Moderate stress4 days0.20
38%Significant depeg8 days0.40
416%Severe depeg16 days0.80
532%Near-collapse32 days1.60
664%Catastrophic64 days3.20

Each tier uses peak-decay memory: the worst event is remembered at full magnitude, then fades exponentially over the tier's half-life.

Production Depeg Events (Test 8.2)

Three real events from defi_materialized.stability_score_mt, one per acceptance-criteria tier (Data Provenance).

AssetReferenceDatetime (UTC)DeviationTierScore
wrapped-bitcoinbitcoin2025-10-10 22:0013.48%3 — Significant depeg0.9673
staked-etherethereum2025-10-10 22:0019.24%4 — Severe depeg0.9570
kelp-dao-restaked-ethethereum2025-09-02 08:0077.98%5 — Near-collapse0.5254

Historical Context

Each detected event corresponds to a documented market incident. The stability scorer classified all three without manual intervention.

Tier 5 — rsETH near-collapse (Sep 2, 2025). Kelp DAO paused all deposits and withdrawals on April 29 after a fee minting contract bug caused excess rsETH to be minted. With on-chain redemptions disabled for months, the only exit was through shallow DEX pools. When the "Red September" crash hit (ETH down 12%, $1.65B liquidated that week), those thin pools were overwhelmed. The 78% deviation reflects a structurally blocked redemption path, not a protocol insolvency.

Tier 4 — stETH/WBTC severe depeg (Oct 10, 2025). President Trump announced 100% tariffs on Chinese imports. The resulting $19B liquidation cascade (largest in crypto history, 1.6M accounts liquidated) collapsed order-book depth by over 90% across venues. 70% of liquidations occurred in a 40-minute window. Binance's price indices relied on internal order books rather than external oracles, amplifying the deviation for wrapped assets. The stETH and WBTC deviations were exchange microstructure failures, not Lido or custodial failures.

Tier 3 — WBTC significant depeg (Oct 10, 2025). Same trigger as above. WBTC's 13.48% deviation was milder than stETH's 19.24% because Bitcoin's deeper liquidity provided more resistance to the cascade. Both recovered within hours as liquidity normalized.

Score Trajectory — Near-Collapse Event

kelp-dao-restaked-eth / ethereum, 2025-09-02. The score craters on impact and barely recovers despite immediate price normalization — the Tier 5 peak-decay penalty (lambda=1.60, half-life=32 days) holds the score down.

Time (UTC)StabilityDeviationScore
2025-09-02 06:000.99260.74%0.9799
2025-09-02 07:000.98721.28%0.9799
2025-09-02 08:000.220277.98%0.5254
2025-09-02 09:000.98541.46%0.5257
2025-09-02 10:000.99420.58%0.5260
2025-09-02 14:000.96713.29%0.5263

Score Trajectory — Severe Depeg Event

staked-ether / ethereum, 2025-10-10. The Tier 4 penalty (lambda=0.80, half-life=16 days) suppresses the score for weeks after the event resolves.

Time (UTC)StabilityDeviationScore
2025-10-10 20:000.95874.13%0.9857
2025-10-10 21:000.95274.73%0.9846
2025-10-10 22:000.807619.24%0.9570
2025-10-10 23:000.96593.41%0.9559
2025-10-11 00:000.95184.82%0.9554
2025-10-11 04:000.95974.03%0.9532

The score does not recover when the price recovers. The peak-decay mechanism ensures severity classification persists proportionally to the magnitude of the deviation.