Status
What exists, what has been checked, and what has not been done yet. The same list a careful reader would ask for before trusting anything.
Chain, Uniswap v4 and tooling verified
The PoolManager was rebuilt from source and matches the chain byte for byte on both networks.
Mainnet swap data collected
Four pools, 735,161 swaps, read from two providers and checked against each other.
Model trained and backtested
It predicts movement better than a one-line rule (0.79 against 1.03). The project's own test for liquidity providers is passed, 4 pools of 4, and the research does not read that as evidence of better returns. The report says why.
A fee memory that moves only when the price does
A swap counts only if it leaves the tick 3 ticks from the last observation, in a later second. Swaps of one wei, flickers and bursts inside a second change nothing. In a search over 480 real states, moving the fee to its cap took a median of 68 ticks over about 21 seconds (2 ticks before).
Python, Solidity and the browser agree bit for bit
1,057 model, 390 policy and 2,111 state-machine vectors, swaps that count and swaps that do not.
Hook, Treasury, oracle and model contracts tested
Foundry tests: unit, fuzz, invariants, native-currency pools, buyback and burn, the Treasury's price limit and the attacks on it.
Deployed on-chain
The on-chain model hash matches the published weights, and 527 swaps replayed through the live pool match the reference fee for fee.
Real swaps replayed through the hook match Python
Swaps cut from the mainnet ETH/USDG pool run through the hook on a local pool manager; every input, model output and fee equals the reference; the browser recomputes them too.
Treasury buys back and burns the project token
Anyone can trigger it, inside a price limit; every call is listed on the Treasury page. Nothing calls it by itself: a keeper script in the repository does.
Wallet flows: swap and liquidity
Through Uniswap's own router and position manager, with ETH or a token on the paying side. scripts/e2e-trade.mjs runs them end to end against a network or a fork.
Shadow of mainnet: the fee Myelin would have charged on four real pools
Read only. Real swaps replayed through the model and policy with the hook's own integer code, tested against the golden vectors, four sequences of 400 real swaps among them.
Documentation
Quick start, how the fee is set, the model, the Treasury, verification, the backtest results, risks and a build guide.
Review of the contracts by the authors
What was found, measured and left open is in docs/SELF_REVIEW.md.
Independent audit
Not done. The authors' review above is not a substitute for one.
A price the Treasury's buyback cannot be moved against
The swap of a call is limited to about 3% above the pool's average price of the last 30 minutes, kept by an oracle. In the tests a sandwich costs the Treasury at most that band. Holding the price for minutes still moves the average, at a cost to whoever does it.
Mainnet deployment
A first deployment on mainnet was withdrawn: it made a token of its own, and the project token will be launched elsewhere. The hook and Treasury will be deployed again on that token (docs/LAUNCH.md).
Updated 2026-10-22.
What can go wrong is on Risks; what was measured is on Backtest results.