What a closed reference market costs you

A perpetual on gold never stops. The market you would hedge it in stops every evening and all weekend. Most venues treat that as a footnote. We treat it as a risk trait.

Every instrument on Openfield carries four genome traits, and the one people skip past is continuity. It answers a question that only becomes urgent at the worst possible moment: if this position goes wrong, is there somewhere to take the other side right now?

For Bitcoin the answer is always yes. Spot and perpetual markets run continuously across dozens of venues, and a liquidation at three in the morning on a Sunday is an ordinary event.

For gold the answer is: it depends what hour it is in Chicago. Our gold perpetual quotes twenty-four hours a day, and it is priced against two crypto venues and a COMEX future. COMEX shuts for an hour every evening and from Friday evening until Sunday evening. During those windows the perpetual keeps trading, keeps marking, and keeps liquidating, against a reference that is not there.

The instrument does not close. The thing that gives it meaning does.

What actually goes wrong

Two distinct problems, often conflated.

The first is the mark. With the future gone, the estimate is running on the remaining sources, whose agreement was partly held together by the arbitrage against that future. Confidence should fall, and it does, because the source count drops and the divergence between the survivors usually widens.

The second is worse and less obvious: the exit. An account being liquidated on Saturday afternoon needs someone to take the position. On a continuous market, the liquidation engine can be patient and the auction usually clears. When the reference has been closed for thirty hours and everyone knows it reopens with a gap, the auction is hunting for a bid that does not want to exist. The insurance fund pays, and the loss is much larger than the model said, for reasons that had nothing to do with volatility.

Hours are not a constant

The first implementation had the sessions hard-coded as one week of hours. It was wrong within a day, because the CME electronic session is not simply Sunday evening to Friday evening. There is a daily maintenance break, the reopening does not line up with the calendar day in any timezone we care about, and the US cash equity session sits inside the futures session as a distinct window with its own liquidity profile.

Hard-coding those boundaries by hand is how you get an off-by-one hour that surfaces once a week, at the exact hour when it matters most. So the calendars moved into their own module, each one a function of a timestamp, and the next state change is found by scanning forward a minute at a time until the answer flips. It is not clever. It is impossible to get wrong by mis-deriving a boundary, which is the property we wanted.

The terminal then draws that answer where a trader will actually see it: a line on the price axis at the reference close, and a countdown to the next open. Not buried in a market-info modal. On the chart, next to the price.

A risk factor a trader has to go looking for is a risk factor they will find out about afterwards.

What we have not built yet

Continuity currently acts as a static trait plus a live factor when the reference is shut. It does not yet do the thing it should eventually do: tighten margin ahead of a known close. The right behaviour is for maintenance margin on gold to begin rising an hour before the Friday close, so that an account which would be unliquidatable over the weekend is either reduced or closed while there is still a market to do it in. Announced in advance, visible in the terminal, and applied to everyone equally.

That is the next piece of work on this part of the engine. Until it ships, the honest description is that we price the closure into leverage and we do not yet price it into margin, and a position carried across a weekend gap is riskier than the numbers currently say.