Routing & splitting

For each swap, chop quotes both DEXs at the full size in parallel, then runs a grid search over possible 50/50 → 99/1 splits. The route with the highest output (after gas) wins.

The decision

Quote DeDust and StonFi in parallel for the full amountIn. Whichever returns a higher amountOut is the single-DEX baseline.
For α ∈ {1/N, 2/N, …, (N-1)/N} (default N=20), quote DeDust at α·amountIn and StonFi at (1−α)·amountIn. Keep the allocation with the largest combined output.
Compare the best split to the baseline. The split wins iff:

text

splitOutput  >  bestSingleOutput  +  extraGasInOutputUnits

extraGasInOutputUnits is the additional gas a 2-leg route consumes (StonFi’s gas + DeDust’s gas − the baseline DEX’s gas) converted to the output asset using the prevailing rate. If the gain doesn’t cover gas, chop returns a single-DEX route.

Why a grid search?

The optimal allocation between two CPMM pools has a closed form, but TON pools include fees, multiple curve types (volatile vs stable), and pool depths that move between probes. A grid is robust against all of that and 20 probes complete in well under a second.

You can pass a smaller slippageBps to tighten min-out limits, but the grid resolution itself isn’t configurable from the public API.

What the data looks like

Live behavior on the TON/USDT pair, May 2026:

Order size	DeDust solo	StonFi solo	Best route	Split share
`10 TON`	24.24 USDT	24.20 USDT	DeDust 1-leg	—
`100 TON`	242.7 USDT	242.2 USDT	DeDust 1-leg	—
`1k TON`	2,415	2,420	2,422 (split)	SF 80% / DD 20%
`10k TON`	22,902	24,040	24,060 (split)	SF 88% / DD 12%
`100k TON`	153,716	225,293	227,121 (split)	SF 88% / DD 12%

Below ~100 TON, splitting’s output gain doesn’t clear the gas cost of a second leg, so chop stays single-DEX. Above that, the depth difference between pools makes a split materially better.

Wallet message limits

TON wallet v4 broadcasts at most 4 outgoing messages per external message — that’s our hard ceiling on number of split legs. Wallet v5 raises it. With two DEXs we’re always at ≤ 2 legs, so this only matters as more DEX adapters are added.