Abstract
A new Solana pair has volume and nobody accountable trading it. wim gives the desk eight operators and one judge. The judge is JEV, a fixed function that reads a coin and answers with one word and three numbers: BUY, SELL or HOLD, a confidence, a trade risk and a rug risk. JEV places no order. Its answer is passed through a gate written in code, and the gate decides. Every fill is a paper fill, sized small, priced with slippage, and marked to the live market. Nothing on this page is a claim about price.
The problem with a trader that is a person
A memecoin desk is usually a wallet with a person behind it, and the person can do anything: ape, panic, size up after a loss, sell into nothing. Nothing about that arrangement can be checked from outside until after it has gone wrong.
The usual answer is to give the desk an agent and let it act. That moves the problem. An agent that can do anything is a degen with better uptime. What a desk needs is a narrow judge: something that can only answer, whose answer follows from numbers anybody can read, and whose answer cannot reach past a set of limits it does not control.
What JEV sees
The desk reads Solana pairs from DexScreener: the top and latest boosted tokens and the latest token profiles, up to 30 tokens, cached for 20 seconds and shared by every visitor. Each pair is reduced to a snapshot. The snapshot is the only thing JEV sees.
b1 = buys1h / (buys1h + sells1h) , b5 = buys5m / (buys5m + sells5m)(1)
L = liquidity / fdv , A = min(1, (volume1h / liquidity) / 1.5)(2)
Alongside these: price, liquidity, pair age in minutes, and the price change over 5 minutes, 1 hour, 6 hours and 24 hours. No wallet, no chat, no holder message is part of the input.
Rug risk
Before JEV is asked anything, the snapshot is scored for the ways a fresh pair usually dies. Each line adds to the score, and the score is clamped between zero and one.
rug risk, additive| condition | adds |
|---|
| liquidity below $60K | up to 0.35 |
| liquidity / fdv below 5% | up to 0.25 |
| pair younger than 2h | up to 0.20 |
| 1h buys below 45% | 0.15 |
| up more than 1000% in 24h | 0.10 |
| down more than 15% in 5m | 0.15 |
One answer
JEV reduces the snapshot to two figures. Momentum weighs recent price change, bent through tanh so that a single wild candle cannot dominate. Flow measures which side is trading.
M = 0.4·tanh(m5/10) + 0.4·tanh(h1/25) + 0.2·tanh(h6/80)(3)
F = 1.2·(b1 − 0.5) + 0.8·(b5 − 0.5)(4)
Each operator has two weights, one on momentum and one on flow. A negative momentum weight makes a contrarian. The weighted sum is the score, and the score is the word.
S = pm·M + pf·F → BUY if S > 0.12 , SELL if S < −0.15 , else HOLD(5)
conf = clamp(0.55 + 0.6·|S| + 0.15·A − 0.2·rug , 0.30 , 0.99)(6)
Confidence rises with conviction and activity and falls with rug risk. With every answer JEV also returns a short list of reasons — liquidity, 1h buy share, 1h change, volume against liquidity — and those reasons are what the stream on the left prints.
The gate
JEV answers. Code decides. Every answer is put through the same gate, in the same order, and ends in one of four verdicts: FILLED, REVIEW, REJECTED or SKIP.
hard limits, enforced in code| limit | value |
|---|
| minimum liquidity | $15K |
| minimum pair age | 30 min |
| minimum liquidity / fdv | 2% |
| maximum rug risk | 0.55 |
| review band below floor | 0.10 |
| exit floor discount | 0.10 |
| daily loss limit | -10% |
A BUY on a coin that fails any hard filter is REJECTED and the failing filter is printed as the reason. A BUY that clears them fills only if confidence meets the operator's floor. Within 0.10 below the floor it is held for REVIEW. Below that it is skipped. A SELL only matters if the operator holds the coin, and exits use a floor 0.10 lower, so leaving is easier than entering.
rule 1, the answer cannot override a limit
A coin that fails a hard filter is never bought, however confident JEV is.
why it holds
The gate runs after JEV and reads JEV's answer only as input. No field in the answer changes a limit, and a rejected answer leaves the operator's book untouched.
Size and fills
Size is not part of the answer. It is computed from the operator's bankroll and the confidence.
size = min(5% · bankroll , $100) · clamp(conf , 0.5 , 1)(7)
slip = min(3% , 2 · size / liquidity + 0.2%)(8)
A buy fills above the market by the slippage in (8), a sell below it. Thin pools cost more, so a trade that looks good on a $20K pool pays for it in the fill. Each operator starts with a $1,000 paper bankroll. Entries are replayed at real DexScreener reference prices from 1, 6 or 24 hours ago and marked to the live price, so every P&L on the desk comes from actual market movement.
rule 2, size is capped twice
No single fill can exceed $100.
why it holds
Equation (7) takes the smaller of 5% of bankroll and $100, then scales it down by confidence. The answer contains no size, so JEV cannot ask for more.
Eight operators
The operators share JEV and the gate. They differ only in their weights, their confidence floor, the reference window they enter from, and how many positions they may hold.
operator parameters| operator | style | momentum | flow | floor | slots |
|---|
| wimnita | momentum | 1.0 | 0.4 | 0.86 | 3 |
| wimman | order flow | 0.3 | 1.0 | 0.85 | 3 |
| wimo | trend 24h | 0.8 | 0.6 | 0.88 | 2 |
| wimie | breakout | 1.3 | 0.2 | 0.84 | 4 |
| wimana | contrarian | -0.9 | 0.5 | 0.87 | 3 |
| wimster | scalper | 0.7 | 0.9 | 0.82 | 4 |
| wima | cautious | 0.6 | 0.6 | 0.92 | 2 |
| wim | degen | 1.1 | 0.8 | 0.80 | 4 |
An operator whose day P&L reaches -10% is marked halted and its avatar stays grey. When an operator fills, the desk cuts across its picture.
The kill switch
rule 3, one switch stops everything
desk.halt(): no new orders, open positions frozen.
why it holds
The switch stops the desk from polling for new decisions. With no new decisions there are no new fills, and open positions are frozen where they are.
What is live
This chapter is here so that nobody has to guess. Live today: the DexScreener feed, the snapshots, rug risk, JEV's answers and reasons, the gate and its verdicts, paper sizing and slippage, and P&L marked to real prices.
Not live: wallets, signing, and real orders. No transaction is sent. The desk shows paper fills because those are the only fills there are.
what none of this says
None of this says a coin goes up. A good paper record means the rules have read the market well so far, and they can stop reading it well on the next pair. Nothing here is advice and nothing is owed to a holder. What the page claims is narrow and checkable: the answer follows from numbers anybody can read, the gate runs after the answer, and size is set by limits the answer does not touch.
notes
- a. Market data is read from the public DexScreener API. Market cap is reported fdv for the pair.
- b. JEV is deterministic code, not a language model. The same snapshot gives the same answer.
- c. The desk refreshes decisions every 2 seconds and the market feed every 15 seconds.
- d. The reasons beside a decision are generated from the snapshot. They are not edited.