The proof behind the numbers.
Nothing here is an estimate. Every number comes from records on the public test networks, and you can open each one yourself. Where a result has a weak spot, we say so.
How long from a swap to the trade that answers it?
We ran it 73 times. All 73 trades arrived and 0 went missing. 71 were completed, and 2 were cancelled on arrival because the prices had already moved back.
How we timed itEvery block on Base Sepolia carries a timestamp. We subtract the timestamp of the block with the swap from the timestamp of the block with the trade.
How preciseThose timestamps move in steps of 0.2 seconds, so every result is a multiple of that.
Weak spotMost runs come from one test script that opens the same price gap again and again on a quiet test network, so they show one kind of workload. The brightest bar is the typical time.
Same trades, far fewer wasted ones.
Our contract checks whether a trade is worth it before sending it. To see what that check is worth, we built a second copy without it, which sends every possible trade, and put both through the same 12 price changes.
| 12 price changes each | Checks first | No check |
|---|---|---|
| Price changes looked at | 18 | 18 |
| Trades sent | 3 | 12 |
| Trades completed | 3 | 3 |
| Trades cancelled on arrival (wasted) | 0 | 9 |
| Fees paid on Base Sepolia, in ETH | 0.0000049 | 0.0000113 |
| Fees paid on Reactive, in lREACT | 0.293 | 0.307 |
| Time taken in this run, typical / slowest | 1.8 / 2.0 s | 1.6 / 2.0 s |
What we foundBoth copies made the same 3 trades. The copy without the check sent 9 more that were cancelled on arrival, so checking first cut the fees paid on Base Sepolia by 56%.
What it does not saveThe Reactive network charges for every price change it looks at, traded or not. That cost was nearly the same for both copies, only 5% apart.
Weak spotThis is one run of 12 steps, and the two copies did not start from exactly the same pool sizes.
One example of each way it can end.
Each row is something that really happened. Click any link to see the record on the public blockchain explorer.
| What happened | Time taken | |||
|---|---|---|---|---|
| The trade is made | 0x5be4…3a46 ↗ | CallbackRequested 0x695e…3a15 ↗ | Executed 0xe34f…be88 ↗ | 1.8 seconds |
| Prices move back first, so the trade is cancelled | 0x4db6…6290 ↗ | CallbackRequested 0x6b8f…de97 ↗ | SkippedMarginGone 0xc080…642b ↗ | 1.2 seconds |
| The gap is too small, so nothing is sent | 0xe8b4…0b36 ↗ | MarginBelowThreshold 0x2b13…41aa ↗ | Nothing was sent | No trade |
Not on the markets we tested. Here is why.
We took three days of real trading (11–13 November 2026) and, for 10 token pairs on Base from WETH/USDC to small meme coins, compared the two busiest pools on Uniswap and Aerodrome: how often did their prices drift far enough apart to be worth trading, and for how long?
Other traders closed almost every gap inside the same block, and none was still open two seconds later. On small coins the prices drift further apart, but never by more than the pools' own fees (up to 1.25%), so there was nothing to gain.
Then we compared the price of ETH on Base and on Unichain, where no one can trade both sides in one block. The two chains stayed 1.6 basis points apart at the median, and of 150 gaps worth trading, only 1 was still open two seconds later. This project proves the method and its speed. It does not beat the competition on these markets.
How we did the replay ↗The contracts used for the comparison.
These belong to the copy without the first check, from section 2. The main contracts are listed on the home page.
| Contract | Chain | Address |
|---|---|---|
| ArbReactive, no filter | 0x82DEA86F73edF0CC76cd2Ff4Ae8612380e63CE6F ↗ | |
| ArbExecutor | 0x9429334cBda1E85Cd99691aE4f0BBca2786d597C ↗ | |
| Pool A | 0x97898610adD7F402C9D6612e5cF69e395eba7AC0 ↗ | |
| Pool B | 0xE4140Ea6a6152F6F0cf02D13a56B7A7F60781105 ↗ |