On Monday, Binance lists USDBRLUSDT: a perpetual future on the US dollar against the Brazilian real, settled in USDT, trading every hour of every day, at up to 100x leverage. Foreign exchange is the largest market that exists, roughly $9.6 trillion a day, and it shuts on Friday afternoon in New York until Sunday evening. A contract that trades continuously on an asset that does not is a genuinely hard problem, and Binance has solved it in one of the only two ways anybody could. The solution is disclosed in advance and almost nobody trading the thing will read it.
TL;DR
- Binance’s USDBRLUSDT perpetual goes live 21 September 2026, settled in USDT, up to 100x leverage, trading 24/7 against an FX market open five days a week.
- It runs two pricing methods: Standard mode during FX hours (a weighted index from third-party data vendors) and Orderbook EWMA mode from Friday 17:00 ET to Sunday 17:00 ET (an exponentially weighted moving average of Binance’s own orderbook).
- On weekdays the index is external, so Binance is a price taker. At weekends the input to the liquidation engine is produced by the venue holding your position, and Binance claims 53% to 59% of weekend volume across five major venues.
- There was no better option: a frozen Friday index would be a stale-oracle arbitrage. The point is that 24/7 trading on a five-day asset does not remove the gap, it relocates it into the pricing methodology.
- The gaming parallel: every provably fair page grades the mechanism and silently assumes it is the same object at 4am on a bank holiday as at peak. Satoshie’s Chainlink VRF path has no off-hours branch, no fallback mode and no holiday calendar.
Two prices wearing one ticker
Here is the mechanism, in Binance’s own words. During regular FX trading hours, Sunday 17:00 ET to Friday 17:00 ET, “the price index is updated every second as a weighted average of constituent prices from third-party data vendors”. During weekends and public holidays, Friday 17:00 ET to Sunday 17:00 ET, “the system transitions to Orderbook EWMA mode, using an exponentially weighted moving average of orderbook prices to maintain fair and orderly pricing”.
Read that twice, because the second sentence does something the first does not. In Standard mode the number estimates an event happening elsewhere: banks and funds transacting dollars for reais in a market Binance does not own. In Orderbook EWMA mode the number summarises what is happening on Binance.
The ticker does not change, the chart does not break, and the liquidation engine does not pause to mention it is now consulting a different kind of oracle. Your margin is denominated in one measurement regime on Friday afternoon and a structurally different one on Saturday morning, and your position has no opinion about the distinction.
The half everybody asks about is accuracy: is the price right, is the feed manipulated. Good questions, and they have answers. The half nobody asks sits underneath: accurate to what? A price is a claim about something, and when the something stops existing for 48 hours the claim has to be about something else. The question worth asking is what got substituted.
You cannot discover a price in a gap
Shunyet Jan, Binance’s Head of Exchange and Trading, framed the launch this way: “FX is one of the world’s most foundational markets, we’re well positioned to extend continuous price discovery into the gaps.”
Price discovery is a specific thing: people who want to transact reveal, by transacting, what the thing is worth. It requires a market, and in a gap there is no market, which is what the word gap means. What you can do in a gap is collect the opinions of everyone still holding leverage at 2am on a Saturday and call the average a price.
That is not worthless, and a venue that lets you hedge a real exposure on a Sunday is providing something genuine. But it is a different category of fact from a weekday FX print, in the way a poll is a different category of fact from an election. Binance has not extended price discovery into the gaps. It has built a market in the gaps and named its output after the market that is closed.
On weekdays the index protects you from Binance
The load-bearing property of Standard mode is that it is exogenous: Binance’s own orderbook is not the reference. If something deranged happens on Binance at 3pm on a Wednesday, a fat finger, a thin book, a cascade, the third-party index carries on reporting what the rest of a $9.6 trillion market thinks. The external feed is the circuit breaker, the reason a venue-side accident does not automatically become a customer-side liquidation.
Orderbook EWMA mode removes that circuit breaker for two days out of seven. A liquidation prints a trade, the trade moves the average, the average moves the mark, the mark triggers the next liquidation. At 100x leverage, the move required to end a position is one per cent. On a weekday that loop is broken open by the rest of the world voting. At the weekend the loop is closed, and it is closed by design.
Nor is this theoretical. Binance’s own announcement claims its equity perpetuals took roughly 53% of combined weekend volume across five major venues for AI-linked contracts and roughly 59% for the most exposed semiconductor ones, leading the second venue by three to four times. That is a boast about market share and also a description of a feedback loop: at the weekend, the dominant input to the number that liquidates you is the venue’s own flow. Binance is not hiding this. It is the product.
There is a corollary. On a weekday, inflated orderbook activity is a vanity metric that lies to a listings page. Under a mark derived from that same orderbook, volume stops being a metric and becomes an input.
The honest part: there was no good option
It would be easy and wrong to write this up as negligence. Consider the alternative: freeze Friday’s index and carry it through the weekend, and you have a number everyone knows is stale and nobody can update, which is the oldest exploit in this industry. Between a self-referential price and a frozen one, the self-referential price is the better engineering choice, and publishing the methodology in advance is more than most venues bother with.
Which is the actual finding, and it is not really about Binance. Making a five-day asset trade seven days does not eliminate the closure. It moves the closure out of the trading calendar, where it is obvious and everybody plans around it, and into the pricing methodology, where it is a mode transition in a document. The gap did not close. It stopped being visible.
The precise product description is not “FX, now around the clock”. It is “foreign exchange for five days, and a Binance-internal proxy for the other two”. Both are tradeable instruments. Only one of them is foreign exchange, and the contract is not going to tell you which one you are in.
Whose public holiday?
The specification says weekends “and public holidays”. Whose? A USD/BRL contract touches two national calendars that do not agree, and Brazil observes days New York does not. The mode your liquidation engine runs in therefore depends on a list of bank holidays that is not in the contract, cannot be derived from the chart, and is maintained by whoever maintains the configuration.
Binance’s own risk disclaimer says the quiet part out loud: TradFi perps “are subject to high market risk and price volatility (particularly outside traditional market hours)”. The parenthesis is the whole post.
We have twice covered prices that cannot be checked, and this is neither. The pre-IPO perpetuals piece was about a company that has never traded, so no spot has ever existed and the mark is a venue methodology permanently. The exit nobody chose was about the liquidation engine’s own parameters. Here the referent is the deepest market on the planet, switched off and back on by a timer while the contract referencing it keeps running under the same name.
Every fairness page assumes the mechanism has no off-hours
Every provably fair page in the industry grades one thing: the mechanism. Server seed, client seed, nonce, hash commitment, published proof. All of it answers “can the operator change the outcome”, and all of it is written as though the mechanism were a constant. Nobody asks whether the guarantee has a weekend.
Plenty of casino mechanics quietly do. Pooled jackpots depend on how many people are playing, peer-matched tables need a counterparty, dynamic odds move with exposure. A game that behaves one way at peak Saturday and another at 4am on a bank holiday is functionally two products sharing a name, exactly like a ticker that changes what it measures at 17:00 ET on a Friday. The fairness proof verifies correctly in both, because the proof was only ever about the mechanism, and the mechanism was never the part that changed.
So the unasked half, in our own industry: is the thing that decides my outcome produced outside this building, or by this building’s own traffic?
What we can actually claim
Satoshie’s answer is narrow and structural rather than a promise. The randomness that picks a raffle winner or resolves a coinflip comes from a Chainlink VRF coordinator whose address is fixed in deployed code, readable before you stake anything, and the returned value arrives with a proof the contract verifies on-chain before the payout callback may act. There is exactly one path. No second methodology that engages when the book is thin, no fallback oracle, no holiday calendar, no moving average of our own activity standing in for a number that failed to arrive.
A quiet Tuesday and a busy Saturday produce the same machine. Raffle odds are arithmetic over tickets minted, readable before you buy, so low participation changes your odds visibly, as a number you can see, rather than changing the method behind a number you cannot. That is the whole argument: it is fine for conditions to vary, and not fine for the definition to vary without saying so.
The limits, honestly
Three, and the first stings. Participation is still other people, and other people are not a protocol parameter. A thin round is a real problem for us as for anyone, and the standard cure, seeding the pot with house funds, reintroduces a version of the disease it treats. A VRF proof says nothing about how many people showed up.
Second, this is not an argument against FX perpetuals or against Binance’s engineering. A disclosed dual-mode methodology beats an undisclosed one and the choice they made is defensible. The objection is to the verb in the announcement, not the mechanism in the documentation.
Third, we run ordinary infrastructure that does have off-hours behaviour: Base’s sequencer, our VRF subscription balance, gas markets. Those inherit every criticism above. The narrow claim is only that the thing deciding who wins has no second mode.
Three questions worth asking any venue
- Does the number that closes my position have more than one definition, and what event switches between them?
- When the definition switches, is the new input produced outside the venue holding my money, or by it?
- Is the switch discoverable from the contract itself, or does it live in a calendar somebody else maintains?
Nobody has to break a rule for this to cost you. On Saturday the signs are still lit and the prices are still quoted, and every shutter on the street is down. A market that is closed is not a market at a different price. It is not a market. Everything quoted in the gap is a number about the people still standing in it, which is fine, as long as that is the thing you thought you were trading.
📷 Photo by Khachik Simonian on Unsplash
