Would StableLens have caught it? — backtest report
The 12 most-cited stablecoin / yield-product failures, scored by the StableLens v1.1.0 risk model against curated public-data snapshots from before each event. Honest — where the v1.0 model would have missed a failure, we say so and name the U-2 dimension that closes the gap.
IndependenceScored from public on-chain + regulatory data only. No issuer-provided data. No issuer compensation. See /methodology.
Non-investment-grade verdict ≥ 30d before failure
12/ 12
Model issued a BB grade or worse (≤ B at first alert) at least 30 days before the failure event. Signals elevated risk — allocators should reduce exposure or apply scrutiny.
Sub-investment-grade verdict ≥ 30d before failure
2/ 12
Stricter bar: model issued a CCC grade or worse at least 30 days before failure. The model wasn't just flagging elevated risk — it was forecasting material impairment.
Flagged late (1-29d)
0/ 12
Sub-IG late (1-29d)
0/ 12
Missed (v1.0 gap)
0/ 12
UST / Terra-LUNA
May 2022 · loss $40.00B
Caught early (sub-IG ≥ 30d)Final grade: B
UST held its peg via mint/burn arbitrage against LUNA (its own sister token = endogenous collateral). When confidence broke, redeeming UST minted more LUNA, crashing LUNA, which destroyed the backing — reflexive death spiral.
Model issued B grade 90 days before the failure event.
Backed by an off-chain fund manager (Stream). When Stream reported ~$93M in losses, xUSD — whose backing was unverifiable on-chain — collapsed to ~$0.23.
Model issued B grade 30 days before the failure event.
