bankruptcy price and liquidation price are not the same line

liquidation price is the mark-price threshold where the venue’s liquidation engine takes control of your position. bankruptcy price is the level where the position’s allocated margin has been fully consumed. for an isolated long, bankruptcy sits below liquidation. the difference is deliberate: the engine needs price room to close the position before its equity falls below zero and leaves a deficit.

maintenance margin creates that room. under the isolated approximation, a long’s liquidation price includes MMR, while bankruptcy price removes it. the resulting gap is not spare margin you can use after liquidation; control has already passed to the engine. it is an execution buffer intended to let the engine close above bankruptcy, with the insurance fund behind the process if the tape runs through that level.

for an isolated USDT-M long, the liqsiren approximation is liq = entry × (1 − 1/leverage + MMR). bankruptcy price is entry × (1 − 1/leverage). using the worked-example entry of $97,400, 20x bankruptcy is $97,400 × (1 − 1/20) = $92,530. at 100x it is $97,400 × (1 − 1/100) = $96,426.

the liq-to-bankruptcy buffer equals entry × MMR while the position stays inside the same maintenance bracket. Binance BTC uses 0.4% MMR at the $10,000 worked-example notional, giving $97,400 × 0.4% = $389.60. Bybit uses 0.33%, giving $97,400 × 0.33% = $321.42. these are estimates — exchanges compute exact liq from live margin; always double-check on your venue.

not within the same MMR bracket under this isolated approximation. leverage compresses the full adverse move from entry to bankruptcy, but the liq-to-bankruptcy gap remains entry × MMR. at the $97,400 BTC entry, Binance keeps a $389.60 buffer at both 20x and 100x; Bybit keeps $321.42. a bracket change, live fee or funding accrual, added margin, or another venue-specific adjustment can move the exact figures.

what shrinks sharply is your pre-takeover distance. the Binance 20x long has $4,480.40, or 4.6%, between entry and estimated liq; at 100x it has $584.40, or 0.6%. Bybit moves from $4,548.58, or 4.67%, at 20x to $652.58, or 0.67%, at 100x. once takeover occurs, only the MMR-created buffer remains before bankruptcy.

MMR is bracket-dependent, so notional can change the takeover buffer without any leverage change. Binance BTC begins at 0.4% MMR up to $300,000 notional, then moves through 0.5%, 0.65%, 1% and 2% in the first five listed brackets. Bybit BTC starts at 0.33% up to $300,000, followed by 0.5%, 0.56%, 0.63% and 0.67%.

a larger MMR moves a long’s liquidation price upward while bankruptcy price remains set by entry and leverage in the simplified isolated calculation. that gives the engine more price distance between takeover and zero equity, but it also liquidates you earlier. only the first five tiers are listed here. beyond the last listed bracket, liqsiren’s fallback slightly underestimates MMR, so large notionals require a venue-side check.

if the liquidation engine gets a fill above bankruptcy on a long, the MMR buffer has done its job: takeover happened before position equity reached zero. if the book gaps, thins or wicks through bankruptcy before enough size can be closed, execution can land below the zero-equity line. the resulting shortfall is no longer covered by the liquidated position’s own isolated margin.

the insurance fund is the next loss-absorption layer. if that layer cannot absorb the remaining deficit, the venue can proceed to auto-deleveraging, reducing opposing profitable positions under its venue-specific priority process. exact insurance-fund balances and ADL ordering are not supplied here and vary by venue. bankruptcy price therefore marks a real handoff point between position equity, liquidation execution and shared backstop mechanisms.

compare the venue’s mark with the liquidation estimate, not bankruptcy price, when reconstructing why control passed to the engine. bankruptcy is the zero-equity reference used after takeover; it is not the initial liquidation trigger. last-traded price can print somewhere else because mark and last are different series. no mark-price formula weights, constituent counts or update frequencies are assumed here.

stops, take-profits and conditional orders need a separate trigger check. trigger type is selectable and varies by product and interface, so verify the selection on the account rather than assuming a default. a stop watching one price series and liquidation watching mark can produce a sequence where the engine acts before the stop condition is observed on the series you were viewing.

1. record the venue, contract, side, margin mode, entry, notional and leverage.
2. capture mark, last and the selected conditional-order trigger series separately.
3. identify the active MMR bracket rather than applying the first tier to every notional.
4. calculate both takeover and bankruptcy levels, then measure the price gap between them.
5. include funding, fees and margin changes when reconciling the estimate with the venue’s live figure.

yes, on two venues. OKX publishes forced liquidations on a public endpoint that carries bkPx, the bankruptcy price of the position, on every print; that endpoint was probed on 2026-09-23 and answered. Bybit publishes one too: its allLiquidation stream carries p, documented as the bankruptcy price, on every print — the field table was read on 2026-09-23. Binance forceOrder carries no such field and Hyperliquid publishes no liquidation stream at all, which is why liqsiren stores bankruptcy price in a nullable column where NULL means the venue publishes none. that makes bankruptcy price a directly recordable field rather than only a theoretical output from the isolated formula.

liqsiren’s own tape carries OKX rows only from 2026-09-23, the day that collector was deployed; there is no OKX history before that date and none was backfilled, so a window opening earlier holds none. Bybit rows also begin on 2026-09-23: until that day the collector read Bybit’s allLiquidation frames against the wrong schema and dropped every one, and nothing from before the fix was backfilled. every Bybit row since carries the venue’s bankruptcy price, so the tape holds venue-published bankruptcy prices from two venues — but only from that date, not as a longer history.

the tape records Binance and Hyperliquid prints from 2026-09-19, and Binance throttles its forceOrder stream, so no single public source sees every liquidation on every venue. liqsiren can record available prints and calculate an estimated liq from real brackets, but exact takeover remains a live-margin result computed by the exchange.