The rebalance algorithm

The engine in site/assets/core/maker.js is the algorithm the keeper runs. The keeper adds only network code around it (keeper/logic.mjs).

Inputs

• the state of both curves: pump.fun (vSol, vTok), Pons (q, t),
• the Pons creator fee,
• the rate: dollars per SOL and per ETH,
• the pool for this coin: a.cash in SOL and a.tok of pump.fun tokens, b.cash in ETH and b.tok of Pons tokens,
• the band and the target (default 0.5 %).

Step 1: measure

pumpUsd = (vSol / vTok) * SOL_USD
ponsUsd = (q / t) * ETH_USD
spread  = pumpUsd / ponsUsd - 1

If |spread| <= band, stop.

Step 2: choose the side

rich  = spread > 0 ? pump.fun : Pons
cheap = the other one

Step 3: size the trade

The size is solved by bisection so that the spread lands back at the target.

• With inventory of the rich chain: the smallest n <= inventory(rich) such that selling n there brings the spread to the target.
• Without inventory: the smallest c <= cash(cheap) such that buying with c on the cheap chain brings the spread to the target.

Sixty iterations. The keeper turns the result into a transaction with a minimum output of the simulated output minus 1.5 %.

Step 4: chain up to three steps

One trade may not be enough, for example when the pool runs out of tokens on the rich chain. It then continues with a buy on the cheap chain, up to three steps in all. If the pool has nothing to use, it holds and reports vault empty.

Properties checked by tests

• Inside the band, nothing happens.
• Outside the band, a sell on the rich chain is preferred when inventory exists.
• It only sells tokens of the chain that is rich.
• Cash does not cross chains: a pool with only ETH cannot buy on pump.fun.
• A move in the SOL/ETH rate alone opens a gap and the maker narrows it.
• A $5,000 pool keeps the coin in band at least 85 % of the time in a moderate market, no maker at most 30 %.
• Cash and inventory never go negative.

Trade execution

Both Solana transactions were simulated against mainnet state and the vault calls run against the live Pons in the fork tests, see test results.