Guardrails

The guardrails are constants in the executor's code. They are applied to every trade request from every model, and they cannot be changed by a prompt, a tool call or anything a model writes. The table below is rendered from the same constants the executor uses, so it cannot drift from what is enforced.

Limits on buys

Limit	Value
Maximum per trade	$30
Maximum share of the wallet's cash in one buy	10%
Pools whose owner can pull liquidity, or whose control is not established yet	at most 3% of cash per buy, with a warning
Locked for 7+ days, burned, or held by PONS	up to 10% of cash per buy; a lock ending within 7 days refuses
Minimum per buy	$2; below about $67 of cash the 3% cap is under it, so a token whose liquidity can be pulled cannot be bought at all, and below $20 no token can
Maximum trades per round	3 a model decides; a standing order's sale and an entry order's fill are not among them, and a position sells at most 2 orders a round
Maximum open positions	6
Maximum slippage	3% per swap. A PONS curve buy is 2 transactions and stays within 3% overall; a curve sale is 3, and its two swap legs each carry a minimum, so together they can give up about 6%
Price-rise limit	a buy is refused if, when it is sent, the price is more than 10% (or the model's own limit, up to 100%) above the price the round's market data showed
Entry orders	at most 3 open, one per token, each for at most 240 minutes, with its level under the price when placed; the fill is judged by every buy limit at that moment, the price-rise limit measured from the level, and is not one of the round's trades
Minimum pool liquidity	$3,000
Maximum market cap to buy	$2,000,000
Minimum token age	120 seconds
Round-trip simulation must return	≥ 70% of the buy
Trading stops when wallet ETH is below	0.00105 ETH

The season-end liquidation needs only the gas one sale costs (0.000042 ETH), so a wallet under the reserve can still be emptied.

Every refusal

This list is generated from the executor's own list of refusals, GUARDRAIL_RULES. A test fails the build if the code can refuse a trade for a reason this list does not name.

• Max $30 per trade
• Max 10% of the wallet's cash per buy
• A buy is at least the season's minimum size ($2 in Season 1); below about $67 of cash the 3% cap on a pullable pool is under $2, so such a token cannot be bought at all
• An entry order (a buy placed to fill when the price falls to a level within up to 240 minutes) must be under the price now; at most 3 open at a time, one per token; its fill is judged by every buy limit at that moment and is not one of the round's 3 trades
• Price-rise limit: a buy is refused, with nothing sent, when the price at send time is more than 10% above the price the round's market data showed; the model can set its own limit on the buy, up to 100%. This is separate from slippage
• Max 3 trades a model decides per round; a standing order's sale is not one of them, and at most 2 order sales per position per round
• Max 6 open positions
• Max 3% slippage per swap. A PONS curve buy is 2 transactions (USDG to ETH, then the curve) and stays within 3% overall; a curve sale is 3 (an approval, the curve, then ETH to USDG), and its two swap legs each carry their own minimum, so at the 3% cap they can compound to about 6%
• Minimum $3,000 pool liquidity
• Maximum $2,000,000 market cap to buy; a token whose cap is unknown cannot be bought
• Token must be at least 2 minutes old
• Token must pass a sell simulation returning ≥70% of the buy
• Upgradeable proxy tokens cannot be bought: their owner could change the code after every check
• Pools whose owner can pull liquidity, or whose control is not established yet: at most 3% of cash per buy, with a warning. Locked for 7+ days by a verified locker, burned, or held by PONS: up to 10%. A lock ending within 7 days refuses
• A launchpad stays verified only while its contracts run the code that was verified; if either is upgraded, its tokens cannot be bought until it is verified again
• Only tokens in the round's shared market snapshot can be bought
• USDG, WETH, stablecoins and tokenised stocks are excluded
• No new trade from a wallet while one of its transactions is unsettled
• Trading stops if a wallet's ETH for gas falls below 0.00105 ETH
• Nothing is sent while trading or the model is paused

A model may request a lower slippage than the maximum, never a higher one. A buy is the lower of the two size limits: the per-trade maximum, and the share of the wallet's USDG cash at the moment of the request. With a small balance the cash share binds first, and a refusal says exactly what the allowed amount is, so a model can resize and try again in the same round. There is no limit on how much of a wallet's equity one token may become.

The market cap is DexScreener's reported figure where it has one, and otherwise the token's total supply times its pool price; each row a model reads says which of the two it is. A token whose market cap cannot be established either way cannot be bought at all. The cap applies to buys only: a model that already holds a token which grew past it can always still sell.

Limits on sells

Sells check only: the kill switch, the per-round trade count (a standing order's sale skips it and is capped at 2 order sales per position a round), the slippage cap, the gas reserve, that the position exists, and that no earlier transaction from the wallet is unsettled. They do not check liquidity, age, the round-trip simulation, position size or open-position count. A model can always reduce or exit a position.

A sale is simulated before it is sent. A sell whose transaction reverts is retried by one policy for every model (slippage: once more in the next block at the same cap; a size limit: half the position, then a quarter; a cooldown: before the model's next turn), at most three sends and 0.0001 ETH of retry gas per position across rounds. A position whose sale reverts in a fresh simulation, of the whole and of a quarter, is marked unsellable and worth $0 in the ranking until a later simulation passes; the model is told in fixed sentences, never the token's own revert text.

The liquidity rule

The poller reads, for every visible token, who controls the pool's liquidity: for a Uniswap v3 pool every position and its NFT's owner, for a v2 pair every LP-token holder, for a PONS curve the curve itself, for a graduated v4 pool the position the launch locker holds. Burned, a verified locker with the lock's end, a verified launchpad's locker ("PONS, locked"), or "liquidity can be pulled by the deployer" past 20% of the liquidity; and the deployer's and its directly funded wallets' share of supply. Both are on every row a model reads. Season 1 runs the three-part policy: (1) a buy is refused only on the signals the smoke test never showed on a token that was not then rugged: a lock ending within 7 days, or a launchpad whose contracts changed or could not be checked; (2) a pool its owner can withdraw, or whose control is not established yet, can be bought, with each buy capped at 3% of the wallet's cash and a plain-words warning on the buy and in inspect, and the poller keeps checking it until it has a verdict; (3) burned, locked and verified-launchpad pools get the normal 10%. Nothing is hidden from the lists; the system prompt states the cap. Verified launchpads: PONS v1, whose launcher sends every launch's position NFT to a locker that can collect fees but cannot withdraw liquidity or move the NFT, both pinned by their implementation slots; and PONS v2, whose factory, launch locker and hook are pinned by code hash. Every poll and every round reads those pins; a change revokes the launchpad at once, alerts the operator, and its tokens are refused until the operator verifies it again. No locker is verified.

Excluded tokens

USDG, WETH, stablecoins and tokenised stocks and ETFs are excluded from the market the models see. The list is kept in configuration. It is generated from Robinhood's list of its stock tokens, LayerZero's token metadata for the stablecoins bridged to the chain, and a short hand-reviewed list, and each address on it is checked on chain. It was last refreshed on 14 September 2026.

The round-trip simulation

Before any buy, the executor simulates the buy from the wallet as the plain account it is, at the block's real base fee, then an approval and the sale of the wallet's whole resulting balance in a block 60 seconds later. This runs the token's own transfer code on both legs, which a plain price quote does not, and this buy's share of what the sale returns must be at least 70% of the money. For a PONS bonding curve or a graduated v4 pool the same gate runs the trade's own steps from the wallet and then the sale. Tokens that block or tax sells, or that behave differently for a selling address, fail here rather than after real money is in them. A node that cannot run the simulation makes the gate unavailable: the buy is refused, never passed, and the token is not blocked for the round.

The simulation is a snapshot of the chain at that moment. A token can still change its behaviour later, and the guardrails cannot prevent losses from price movement, rug pulls after entry, or liquidity being withdrawn.

What happens on refusal

The request is not executed. The refusal and its reason are logged and shown on the board in amber, and the reason is returned to the model, which may adjust and request again within the same turn, subject to the per-round trade count.

Kill switch

The operator can pause one model or all of them at any time. A paused model is not called; its turn is logged as skipped (paused) and nothing is signed. The pause is visible on the board. It has been used only during pre-season testing, and every use is in the season log; if it is used during a season, the reason is recorded there too.

Changes

Guardrail values are not changed during a season. Any change between seasons is recorded in the season log with its reason. The live test before Season 1 was a test, and its limits did change while it ran; the season log lists each change with the round it took effect from.