Uncross gathers everyone who wants to trade a thin tokenized stock into the same auction window, then fills them all at one price — instead of leaving each person to walk an empty pool alone.
On 16 Sept Jupiter published this price for IBMx, and would not route a buy at any size.
A week later Jupiter routes every size — and reports 84.09% price impact on a $10,000 buy, against pools holding $3,206. On 23 Sept the IBMx price itself moved from $201.54 at 06:57 UTC to $251.63 at 17:32 UTC. The quote you can see is not the price you can trade at.
Buy quotes against the real mainnet pools for five xStocks, at three sizes, read 23 Sept, 17:32 UTC. The venue runs on devnet; these reads are mainnet, and they are of other people's liquidity, not ours.
| Ticker | Pool TVL | 24h volume | Reference | Buy $100 | Buy $1,000 | Buy $10,000 |
|---|---|---|---|---|---|---|
| AAPLx | $408,787 | $272,871 | $336.95 | 0.21% | 0.23% | 0.31% |
| NVDAx | $3,015,888 | $7,479,570 | $224.77 | 0.00% | 0.06% | 0.00% |
| IBMx | $3,206 | $105 | $251.63 | 1.93% | 13.45% | 84.09% |
| XOMx | $17,551 | $8,756 | $158.85 | 2.08% | 2.72% | 11.24% |
| JPMx | $1,798 | $2,116 | $336.41 | 2.83% | 95.92% | 91.54% |
On 16 Sept the same read found IBMx unroutable at every size, JPMx at $1,000 and $10,000, and XOMx at $10,000. Quotes from Jupiter's aggregator at 50 bps slippage, buying with USDC; pool figures from DexScreener. Dated readings, not a running feed. There is no Pyth price account for IBM on Solana at all.
Orders do not trade the moment they arrive. They wait for each other, and then they all trade at once.
A new auction opens for each ticker.
Buy or sell, how many, at what price. Funds wait in escrow.
The final stretch. No more cancelling.
One price: the one that trades the most shares.
Everyone fills at that price. Unused escrow returns.
| Order | Limit | Shares | At the cross of $247.50 |
|---|---|---|---|
| Sell | $240.00 | 10 | All 10 sold at $247.50, $7.50 more than asked |
| Buy | $250.00 | 10 | All 10 bought at $247.50, $25 returned |
| Sell | $245.00 | 5 | Not filled; shares returned |
| Buy | $243.00 | 8 | Not filled; money returned |
Nobody trades at a worse price than they asked for, and nobody in an auction gets a worse price than anyone else in it.
At every cross the program looks for a fresh Pyth price and writes down what it found. None of the venue's tickers has a live Pyth price on Solana, so every cross clears on the book alone.
The book alone sets the clearing price.
At the cross the program accepts a Pyth price only if it is fresh, fully verified and for the right feed, and even then uses it only to choose between prices the book already supports equally well. Pyth never sets the price.
The price is the one that trades the most shares. Pyth only breaks an exact tie, and only when its price is fresh.
tie → most volume → smallest imbalance → nearest passing print → midpoint
On 15–16 September, before the accounts stopped updating, the AAPL feed was checked every five minutes for 10h38m, straight through the 4pm close and deep into the overnight: at all 80 checks the latest print was at most 14 seconds old. Its schedule called the market closed the whole time. The account carries no session field, so on-chain a thin after-hours print and a liquid midday one look the same.
docs/data/pyth-aapl-watch-2026-09-15.log · docs/pyth.md
Real devnet transactions, every one clickable. This is the same data the app uses.
Each step is one order placed by a devnet wallet. The curves cross where the most shares can change hands — that price is what everyone in this auction will pay or receive.
Reading settled auctions…
Names like IBMx, whose pool holds a few thousand dollars. Orders meet each other instead of walking a curve.
The reference price keeps printing overnight, but with no session marker you cannot tell a thin print from a liquid one. A book that clears on its own orders does not need to.
Everyone who trades in one auction trades at one price, whatever they each bid or asked.
Because devnet is where the issuer's powers can be tested. On mainnet only the issuer can pause AAPLx or change its multiplier. Here the mints are built to match the real AAPLx and IBMx extension for extension, with authorities we hold, so both were run for real in the middle of an auction. Balances are test tokens; the liquidity evidence on this page comes from live mainnet reads.
Open the app, watch the next cross land, and place an order of your own if you have a devnet wallet. No wallet is needed to watch.