Difference in output, in basis points of the aggregator's output; positive means Split Optimizer gives the user more. Size-weighted averages weight each order by its USD size. “Ahead” and “behind” count orders more than 0.05 bps apart.
| Order size | Orders | Ahead | Behind | Median | Size-weighted | Extra output, USD |
|---|
The extra output an aggregator gets by running Split Optimizer next to its own engine and keeping the better quote of the two for each order, after gas, as basis points of the whole volume.
| Order size | Orders | Split Optimizer better | Volume, USD | Extra output, USD | Gain |
|---|
Run Split Optimizer in parallel with your existing router and keep the better of the two quotes for each order. Its optimisation is heuristic, so on some orders it returns less than the aggregator (the “behind” column above). Running both means those orders lose nothing, and the gain is the best-of-both figure.
This is the first version. The heuristics will be tuned on more order data and improved in later versions, which should cut the orders where it falls behind.
Orders. 3,429 sell orders of real size and pair, each taken from an executed CoW Swap trade (linked per order below), from about $13k to $12 million (median $72k).
Same block. On 18 Nov 2026 each order was sent to each aggregator's public quote API. Split Optimizer then quoted the same order on the exact on-chain state that aggregator priced, using a block-by-block recording of the market, so neither side saw later blocks.
Same liquidity. Aggregators were asked to use only the on-chain venue types Split Optimizer covers (Uniswap V2/V3/V4, SushiSwap, PancakeSwap, Curve, Balancer V2/V3, Fluid, Ekubo, ERC-4626 vaults). RFQ and private market makers were excluded on both sides, so the comparison is about routing, not access to private liquidity.
Fees and gas. Aggregator-side fees were added back so only routing is compared. “After gas” charges every route its own gas estimate at that block's base fee, priced in the output token; gas was cheap on this day, so both views are close.
Router. Split Optimizer is a routing engine built on PropellerHeads' open-source Fynd and Tycho, with a proprietary optimisation algorithm that splits large orders across pools and paths and merges flows where routes meet. Median time per quote is about 120 ms (p90 about 150 ms), measured on a single 16-vCPU Google Cloud VM (c4d-highcpu-16).
First version. This is the first version and it will keep improving: latency can be reduced further with more cores and tuning, and throughput can be raised as needed by running more instances, since each one is stateless.
No software needed for the first three steps.
/solve) is available on request, so any order or auction can be compared on current blocks.The manifest lists the SHA-256 of results.csv, routes.jsonl.gz, blocks.json, summary.json and of each aggregator's raw responses. The files are published alongside this page.
| Pair | Size | Block | Split Optimizer | Aggregator | Difference | Pools |
|---|
These are quotes, not executed swaps. Split Optimizer's outputs come from exact simulation of each pool at the quoted block; the aggregators' are their own API quotes.
Left out of the totals and the order list: comparisons where the two quotes differ by 1,000 bps or more, orders under $1,000, and orders the router found no route for. Most of these are tokens the router does not yet cover well (Aave aTokens, USDY, PEPE, ENA and a few others), where the aggregator's quote is far better. Run in parallel, those orders keep the aggregator's quote and lose nothing. Every one is still in results.csv, which the manifest hash covers.