Perpetual markets do not have expiry. They have a funding clock, an oracle, and a liquidation policy. Most ‘unexpected’ losses are those three disagreeing in public.
A dated future converges because expiry forces cash or physical settlement against a known index. A perpetual future never expires, so the market invents a substitute: regular payments between longs and shorts (funding) so that the perpetual tracks the index closely enough to be useful. That is market structure, not a feature toggle.
Oracles decide what ‘closely enough’ means for margin. If the mark used for maintenance is slow, liquidations lag and the book socializes the gap. If the mark is fast and noisy, solvent traders get closed on prints that the underlying never paid. Neither failure is exotic. Both show up whenever a venue copies another venue’s parameters without copying its oracle and insurance design.
The liquidation window is the time between ‘the position is underwater on the venue’s rules’ and ‘the position is gone.’ On-chain, that window is shaped by block time, keeper incentives, and whether close-out is a taker sweep, an auction, or an auto-deleveraging haircut. A desk that only models price paths and ignores keeper economics will be surprised by the path that actually happens.
None of this is a recommendation to trade a named venue. It is the checklist we use when we talk about on-chain trading infrastructure: funding formula, oracle sources and heartbeat, maintenance vs initial margin, liquidation waterfalls, and who eats the residual. If those are not written down, the product is a story about leverage, not a market.
Desk notes on on-chain trading. Not investment, legal, or tax advice; not an offer of securities; not a live quote or a signal. Protocol and product studies live under Selected work.
← All essays