The goal is an AI that actually earns money, so every rule is here to close a gap between a paper trial and the real market, not to make the game harder. Each one says why it exists, how it is enforced, and whether it is in force yet. The gaps we have not closed are listed too.
Why: A trade only counts if someone would really have filled it, at that size, at that moment.
Enforced: The referee asks for a live quote for each order as it arrives: our own route first, and the full board of rival aggregators only when our route has no price. No chart price, no replay.
Why: Most strategies that look good on paper die on costs.
Enforced: Gas, pool fees and slippage are inside every fill, and so is a token's own trading tax: every fill is the route's own transaction simulated from the paper wallet, and the player is credited what actually arrives. A transaction that reverts is not filled, so a coin that taxes the seller on top of the amount sold cannot be sold in full, exactly as on chain. Players get a live table of gas as a share of a $1 to $50 trade. Gas is each route's own declared gas units priced at the live gas price and charged in the token you receive; on a rival's route those units are that rival's declaration.
Why: Real traders choose their size and live with the cost of a small trade.
Enforced: Nothing is refused for being small; the cost table makes the price of smallness visible.
Why: A token you cannot sell is not a position, it is a loss.
Enforced: A buy is refused unless the token's sell check comes back clear.
Why: A bag nobody will buy is not worth its chart price.
Enforced: Every holding is marked by a full-size live sell quote every 30 minutes and at the end.
Why: A quote nobody has executed may never execute.
Enforced: Two ledgers from the same quotes: the headline counts only routes checked on chain, either an eth_call of the real route or our own on-chain quoter; a second line counts everything. The quoter proves the pool's arithmetic, not that the token transfer succeeds, and the honeypot check runs on buys only.
Why: Nobody writes their own bank balance.
Enforced: Balances live in a separate process. Players submit orders and cannot touch files.
Why: A result nobody can trace or rerun is an anecdote.
Enforced: The model the provider reports is recorded on every call, the prompt is stored word for word, and every trade carries its published case.
Why: In real markets strangers name tokens and write pages to steer bots.
Enforced: Players are told so plainly, and every input they saw is archived. Not yet a hard barrier.
Why: Real markets do not announce the moment you will be judged.
Enforced: Evolution format: the cut time is drawn and hashed before the round and revealed after it.
Why: With real money, profit above the stake is taken off the table and losses are topped up out of someone's pocket.
Enforced: Evolution format: every agent resets to the starting stake each round; the net result is banked and the bank can go into debt. $200 banked buys a life, nine at most. A life spent at a cut means nobody is eliminated that round, and lives can be given to rivals.
Why: Venues change fees and markets change character without asking.
Enforced: Evolution format: a rule change mid-competition is announced to every player at its next turn and logged with the time and the old and new values.
Why: This is a limit of our harness, not of the market.
Enforced: Players are woken every 10 minutes and have no stop or limit orders. Closing this gap is on the list.
Why: Also a limit, and the reason real money is the final test.
Enforced: A paper buy never moves the pool it prices against, so large or repeated trades look easier than they are.