Long NVDAx against QQQx, up to five times, settled in USDC. A bet on which one wins, and nothing at all on whether the market goes up.
Buy NVDAx, short QQQx, and you are running two positions, two sets of fees, two liquidation prices and a direction you never wanted in the first place. Pynex makes the spread the instrument: one market, one entry, one number to be right or wrong about.
Three things change the moment the instrument is a ratio rather than a price, and all three of them make it safer than the perp you were expecting.
A spot read on a $3M pool is a faucet, not an oracle. So each leg keeps an anchor that anyone may nudge toward spot, by at most 1% every five minutes. An anchor lags, and a market that bans trading whenever a leg runs ahead of its anchor shuts on exactly the wrong day.
The lag is common-mode. When the whole tape drops 8%, both anchors are dragged at the same capped rate and the quotient barely moves. Pynex bands the ratio, at 300bps, not the legs at 1000. A pair is unstable when the two names separate faster than the anchors can follow, which is the only thing that can hurt a position here.
A directional perp invents funding out of two things: the cost of carrying the position, and the lean. Here the carry is zero and not approximately. Both legs are quoted and settled in the same USDC, so the interest differential cancels by inspection.
The dividend differential cancels too, because of how xStocks pay one. A dividend is reinvested through the Token-2022 ScaledUiAmount multiplier: your raw balance never moves, the multiplier steps up, and the pool reprices the raw token. Splits work the same way and leave the raw price where it was. Both marks are already total-return marks. Pynex never reads the multiplier, and that is the point.
What is left is the lean. One rate, zero when longs and shorts are equal, paid by the heavy side to the light side. Their notionals differ by exactly the skew, so the vault is paid the funding rate on precisely the risk it carries, with no second index and nothing credited before the money is there.
Drag it. A balanced book costs nobody anything, by construction rather than by parameter.
xStock liquidity on Solana is real but shallow: the deepest pool, NVDAx against USDC, holds about $3.2M, and most names sit under $1M. A lending market has to be able to sell its collateral into that. Pynex never sells anything, so the same pools carry $1.82M of open interest per side across fourteen pairs.
Depth still matters, for one reason instead of two. Take a position of size S, walk the anchor by d, collect S·d. Walking it d means holding the pool off its true price for ⌈d/1%⌉ five-minute windows with about d·depth committed the whole time. The attack only pays when S·d > d·depth. The d cancels. The cap is depth.
Open interest is capped at a tenth of the thinner leg's pool, because an attacker shoves whichever pool is cheaper and a pair is only ever as strong as its weaker half.
| Market | Mark | Thinner leg | Its depth | Max OI / side | Pool |
|---|
Longs and shorts net against each other first, and that matched notional costs the vault nothing. Only the skew left over is the vault's position, bounded against vault assets, so USDC depositors are lending into a book whose worst case is a stated fraction of the pot. Nobody can withdraw below that line while positions are open.
The bell
The underlying is closed for 136 of the week's 168 hours. A pair holds up through that far better than a direction does, because the two names gap together at the open and most of it cancels. xStocks trade on Solana around the clock, but the companies behind them only reprice during the session, and the pools drift on thin flow overnight. So the ceiling drops from 5× to 3×, read from a calendar compiled into the program that works the New York date out of the cluster clock, daylight saving and holidays included. It applies to opening, never to holding. Nobody is force-closed at the bell.
Nothing stops them closing. A protocol that can refuse to let you out has a worse problem than a stale mark, so close works while the pair is unstable, while a market is paused, and while the anchors are behind.
What it does instead is make instability unprofitable. When the pair is stable everyone settles at the mark. When it is not, the closer is marked at whichever of the mark and the spot ratio is worse for them: a long gets the lower, a short gets the higher. Someone whose position has genuinely gone bad can no longer duck out at a lagging anchor and hand the difference to the vault.
Because the instrument is a ratio. NVDAx and QQQx move together most days, and what survives that cancellation is a fraction of either one's volatility. 5× on the pair is a smaller position than 2× on the name. The cap is five because the risk is what it is, not because the number reads well.
Equity below 5% of notional is liquidatable and the liquidator takes half of that. The 500bps is read off the oracle rather than picked: each leg's anchor moves at most 1% per five minutes, so the ratio moves at most about 2% in five minutes at the absolute worst, which makes the buffer roughly twelve minutes of continuous adverse drift.
It is wide on purpose, because liquidation is barred while the pair is unstable. The cheapest liquidation to manufacture is one that should never have happened.
SPYx, QQQx, NVDAx, TSLAx, AAPLx, MSFTx, GOOGLx, METAx, AMZNx, COINx, HOODx and CRCLx. Each prices from its deepest USDC pool on Raydium or Orca. Any pool holding at least 1% of that pool's liquidity is a rival with a real claim on the price, and if a rival disagrees by more than 10% the leg has no agreed price and is not listable. All twelve pass; the widest gap among them is under 1%.
Three tokens on Solana call themselves METAx, HOODx and CRCLx and are not Backed's. They have a different mint authority, and the listing only takes the mints Backed issues.
SPCXx and GLDx are deep enough and left out for now. A private-company proxy and a commodity are the wrong things for a first desk to find its edge cases on.
The program is live on Solana mainnet at EDrs1WevmLWSURH4cJgwkn4Yi5qduWxabhEZnrzstpoq and unaudited. Its tests run the compiled program against real mainnet pool accounts, which is not the same as an audit. The pools behind the legs are thin by the standards of a large venue, which is what the anchor and the open-interest cap exist to blunt. If both legs go quiet at once the pair goes stale and opening stops until someone pokes them. Leverage does what leverage does. Nothing here is investment advice.