Liquidation Stress Test
API ▸ How to call thisApply a hypothetical price move to a modeled liquidation map built from recent derivatives snapshots and tracked Hyperliquid positions. The result is an estimate of notional that could be at risk under the scenario—not a list of accounts known to liquidate.
Estimated liquidation clusters by price level
Estimated — not realizedGreen bars = long liquidations (triggered by a price drop). Orange bars = short liquidations (triggered by a squeeze). Hatched, paler bars contain open interest from a venue that publishes no long/short split — the model divided it evenly, so their symmetry is an assumption rather than a measured balanced book. The dashed marker is your target price. Bars are estimated from current open interest, real whale positions, and leverage assumptions — not a settled outcome.
Triggered by exchange
Realized-liquidation context
How the model works (read this)
The reference price is the mean of the latest usable exchange prices. For each tracked Hyperliquid position, the model combines its reported entry and leverage with a 0.9 initial-margin-loss factor to estimate a liquidation level. This is not the venue's account-specific liquidation price: maintenance tiers, cross margin, collateral changes, and account equity are not available.
For the rest of visible open interest, the model infers an average leverage from funding magnitude and distributes that OI across five leverage bands from 0.5× to 1.5× the inferred average. A down scenario crosses modeled long-liquidation levels; an up scenario crosses modeled short-liquidation levels. A venue with no recorded funding rate gets no bands at all rather than a default leverage, and is listed as uncovered under “Venue coverage”.
The long/short split is only shown per side where the venue publishes one, and what it publishes is an account ratio (Binance globalLongShortAccountRatio, Bybit account-ratio) — a headcount, not notional. Hyperliquid publishes no such ratio. For a perpetual, long and short notional are equal by identity, so that venue’s OI is divided evenly by the model; the page then shows one aggregate figure for it rather than two per-side figures, because an even division is an assumption and printing it as “L $x / S $y” would present the assumption as two observations. Bars containing such OI are hatched and labelled “split NOT confirmed”.
Scenario severity is crossed modeled notional divided by measured visible OI: Low <12%, Moderate 12–33%, High 33–66%, Extreme ≥66%. It is a scenario-size label, not a probability forecast. When no venue reported open interest there is no denominator, so severity reads Unknown and the share is blank — it does not fall back to 0% or to “Low”.
Hyperliquid whale positions are also part of Hyperliquid aggregate open interest. Because the two model components are merged without account-level de-duplication, their notional can overlap. Read the total as a stress indicator, not an additive estimate of settled liquidation volume.
Data, freshness & how to read
The endpoint uses the latest derivatives and whale-position snapshots that are no more than 2 hours old and is cached for up to 2 minutes. Exchange coverage depends on which fresh rows are available. Realized-liquidation figures come from a separate persisted forced-liquidation stream and are shown only as recent context; the venues in that sample are named in the panel above from the response itself, not from a fixed list here — which venues are present varies by symbol.
- The bars show all modeled clusters; the KPI counts only clusters crossed by the selected target price.
- “Nearest wall” is the closest modeled cluster, not support or resistance.
- A 0% move should produce zero triggered notional.
- Realized context does not validate the modeled scenario.