Why you got liquidated at a price the chart never touched

Mark price, index price and last price are three different numbers on every Polymarket Perps market, and only one of them can close your position. The chart shows you a different one.

The three prices

Every market on Polymarket Perps publishes all three at once, and the tickers endpoint hands you index, mark, last and mid side by side if you want to watch them diverge in real time.source↗

That last cell is the whole story. Five separate mechanisms — every one of which can cost you money — read mark and none of them read the chart.source↗ Your equity is collateral plus unrealised PnL measured at mark, so the moment mark moves your margin ratio moves with it, whether or not anything traded.source↗

The chart records what other people paid. Mark decides what you are worth.

Where index comes from

Index price is not a Polymarket opinion. It is an aggregation of 3 external providers — Pyth, Chainlink Data Streams and Hyperliquid — combined into one number and republished 5 times a second.source↗

The combination is not a naive mean. Weighted average across the selected feeds, after dropping stale prices and filtering outliers. A frozen feed is far more dangerous than a missing one, because it still looks like data. Filtering outliers handles the other failure mode, where one provider prints something absurd for a second and a half.

Which feeds are in the set is not fixed. It varies by market session, so an equity perp is priced from a different combination of providers at 3am than during the cash session.source↗ The session categories are 5: Regular, Overnight, Weekend, Disrupted, Halted — and with no dedicated feed set for the current category, the overnight set is used.

This is also why a session change is far less dramatic than traders expect. Sessions change which external feed set computes index price and the c3 candidate in mark price, and explicitly not funding, margin and leverage tiers, order matching, liquidation triggers.source↗ The overnight and weekend behaviour has its own page.

The three mark candidates

Mark is not calculated. It is elected. Three candidate prices are computed independently, from different inputs, and mark is whichever one lands in the middle.source↗

Mark = median(C1, C2, C3)

C1 = Index + EMA(Mid − Index)

C2 = median(BestBid, BestAsk, LastTrade)

weighted average of external mark feeds, outliers filtered

Read those three weaknesses together and the design becomes obvious. C1 lags. C2 is manipulable. C3 depends on somebody else’s uptime. There is no fourth candidate without a flaw, because there is no flawless way to price a thing. So instead of hunting for one, Polymarket built three with uncorrelated flaws and let them vote.

C1 smooths the gap between the book’s mid and index with an exponential moving average over a 150-second window — long enough that a single aggressive sweep barely registers, and long enough that it trails a genuine repricing.source↗ C2 takes the median of best bid, best ask and last trade, with stale trades excluded so a print from an hour ago cannot anchor a live valuation. C3 is built from aggregated external mark feeds — a separate feed set from the ones that build index — with a median, an outlier filter and a weighted average on top.

Whatever comes out is then snapped to the market’s tick and rounded to its price precision before it is published, so there is no hidden sub-tick mark that the risk engine uses and you cannot see.source↗

Why a median of three, and not an average

An average of three candidates would let any one of them move mark. Push C2 by ten dollars and the average moves by three and a third. Every candidate is a lever, and the cheapest lever to pull is always the local order book.

A median of three has a property an average does not: the output is always one of the inputs, and specifically the one two other candidates bracket. To move the median you must move at least two of the three candidates in the same direction. Move only one, however violently, and it becomes the highest or lowest of the three — and the median discards it entirely, along with all of your effort.

At least two of three must agree before mark moves. No single signal can move it alone.

And because the three candidates draw on different inputs — a smoothed premium over index, the local book, and an independent external feed set — moving two of them at once means simultaneously manipulating Polymarket’s book and the outside market it references. At that point you are not spoofing a perp; you are moving spot, which is a fundamentally more expensive hobby.

A worked spoof

Arithmetic, not a live reading. Take a market with index at 100.00 and a book that is quietly in line with it. Somebody then sweeps the bids in a thin moment and prints a fill down at 96.00 — a 4.00 hole, 4%, comfortably past the 2.56% adverse move that liquidates a 20x long, if it counted.

Mark lands at 100.01. The wick that would have wiped out a book full of leveraged longs on a venue that margins off last trade does nothing here except leave a mark on the chart. The 150-second EMA in C1 is why the sweep barely touches that candidate at all: it has 150 seconds of history to argue with.

Now run it the other way, which is the case that actually costs people money. Suppose the underlying genuinely gaps down and the external feeds move, but Polymarket’s own book is thin and slow to follow. C1 and C3 both track the real move; C2 lags on a stale local book. Median takes the middle of the three — which now means mark leads the chart. Your margin ratio deteriorates on prices that have not printed locally yet, and if it crosses MarginRatio < 1.0, which means Equity < MM the liquidation engine does not wait for the book to catch up.source↗

When feeds break

Every candidate has a defined failure mode, and every one of them falls back to the same place: Index.source↗

• ▸Lose the local book and C1 and C2 have nothing to smooth or median — they collapse to index.
• ▸Lose the external mark feeds and C3 collapses to index.
• ▸Lose enough of both and mark becomes the median of three identical values, which is index.

That is the whole safety net. In the worst case Polymarket can construct, mark simply tracks the underlying’s fair value — itself a weighted average across multiple providers with stale prices dropped and outliers filtered, so the floor beneath the floor is still not a single point of failure.source↗

Note what does not happen: mark never falls back to the last traded price. There is no degraded state in which the chart starts margining your account.

So the chart lied to you (it didn’t)

You were liquidated at a price the chart never touched because the chart plots the last traded price and liquidation is evaluated against mark. Two different numbers, computed from different inputs. The gap between them is not an error, a lag, or somebody hunting your stop — it is the safety feature you would have wanted five minutes earlier, pointed the other way.

The same asymmetry runs through everything sitting on top of mark. Your stop-loss watches mark, so it can fire on a move the chart never printed, and it can decline to fire on a wick that did.source↗ Polymarket is explicit that triggers watch mark price. Every order type, in full.

The funding premium is measured against index rather than mark: the premium index compares an impact price walked off the book to the index, and mark is what your position is then valued at.source↗ The funding page has that arithmetic in full.

And if your liquidation price itself keeps drifting while you sit there doing nothing, that is a different mechanism again — cross-margin equity, funding accrual and unrealised PnL all push it around. That one has its own page.

Stop reading the chart as though it were your account. The number that closes your position is not the number you have been staring at. The overview explains the rest of the machine.