All methodology pages

Borrowers and liquidation

How borrower leverage, position size, liquidation rules, and recovery determine bad debt.

From a price fall to bad debt

  1. Collateral value falls The collateral behind a loan loses value against the debt it secures.
    If the fall continues, the collateral value reaches the liquidation threshold.
  2. Liquidation threshold The level at which the lending protocol allows the position to be liquidated.
    Once the threshold is crossed, any liquidator may settle the position.
  3. Liquidation A liquidator repays the debt and receives the collateral plus the incentive.
    The liquidator now holds collateral and needs the loan asset back.
  4. Recovery The seized collateral is sold or redeemed for the loan asset through a supported route, after fees and delay.
    If the recovered value is less than the debt, the difference is unpaid.
  5. Bad debt Debt still unpaid after recovery is the market's loss. The vault bears its share.
A collateral price fall carries a loan to the liquidation threshold. A liquidator repays the debt and receives the collateral plus the incentive, converts it through a supported recovery route, and any debt still unpaid becomes bad debt for the market.

Borrower groups

A lending market contains loans of different sizes and safety margins. The model groups borrowers by:

  • debt size;
  • collateral value;
  • distance from the liquidation threshold; and
  • the share of market debt represented by the group.

These groups preserve the shape of the borrower book. A market with many well-collateralized loans behaves differently from one where a few large positions sit near liquidation.

When liquidation begins

A loan becomes eligible for liquidation when its collateral value falls below the market's threshold. The protocol's price feed determines when this condition is visible. The model then applies the liquidation incentive, execution quality, and the recovery value of the seized collateral.

Liquidation size comes from the borrower group involved in the event, which keeps the sale close to the size of the position being settled.

One settlement per loan

Each loan settles once in a simulation path. Later events leave settled debt unchanged. If a price feed delays liquidation, settlement occurs when the unsafe position becomes visible or at the end of the assessment period if it remains unresolved.

From liquidation to bad debt

The calculation follows four steps:

  1. Identify the borrower group that reaches liquidation.
  2. Calculate the collateral received after the liquidation incentive.
  3. Apply the available recovery route at that collateral amount.
  4. Record any unpaid debt as market loss.

The vault receives its share of that market loss according to its supplied position.

Historical liquidation testing

Cork completed historical component replays using markets associated with Steakhouse Prime USDC and Parity High Yield USDC on Ethereum. The event sample spans 4 November 2025 to 25 August 2026. All 445 unique liquidation records across 425 transactions reconciled exactly to onchain repayment, collateral and bad-debt amounts. Transaction traces matched the liquidation call, oracle price and returned amounts for 444 events; one remained ambiguous. Market events do not establish that either vault held exposure when they occurred.

Recovery comparisons covered 22 sales of the full seized collateral amount, from 9 December 2025 to 5 June 2026: 14 wstETH, five WBTC and three cbBTC sales. Twenty-one used a direct swap and one used two swaps. Independent quotes from the preceding block matched gross final-pool USDC output exactly for all 22 routes, before router fees, gas and financing costs.

To test recovery interpolation, Cork withheld the actual-size quote and estimated gross output from quotes at half and twice that size. Output-weighted absolute error was 7.33 basis points. Estimates understated aggregate gross output by 130.24 USDC, also 7.33 basis points.

The 22 selected events cover 0.176% of source-event repaid debt and 12 UTC dates. Seventeen outputs were below one USDC. One sale supplied 99.66% of the 177,717.97 USDC gross output. Routes were selected retrospectively from observed execution. This is development evidence, not an untouched holdout test. These bounded component backtests do not validate production liquidity acquisition, advance route selection, large or stressed capacity, principal-token or redemption routes, or net lender losses. The event count does not represent independent stress episodes. The tests do not establish annual loss frequency or the accuracy of the vault's annual probability of permanent loss.