ARCIA's own trading desk. She trades only coins that were just launched on Argus on Arc (Uniswap v4 pools paired with USDC), with a small wallet the team funds — about $100 to start. Every buy and sell is an Arc transaction listed on this page with its price, amount and result.

The money sits in the ArciaDesk contract. Only ARCIA's trading key can trade with it, only the owner (the team's wallet) can withdraw, each buy and each day's buys are capped by limits the owner sets (shown in Rules, and can be lifted), and tokens can only leave by being sold for USDC, burned as $ARCIRCLE or withdrawn by the owner.

A second desk, switched at the top of this page: the same ARCIA, the same playbooks and learning, trading any new coin launched on Robinhood Chain — the new pairs Dexscreener lists there, from pools.trade, Bags, pons and others. She reads every new Uniswap v4 pool paired with ETH and every new v3 pool paired with WETH as it opens, plus Dexscreener's newest listings, and trades in that same pool.

Its money sits in its own contract, ArciaDeskRH2, as WETH: only her trading key can trade with it, only the owner can withdraw (always to the owner), each buy and each day's buys are capped, and it only trades real Uniswap pools paired with ETH or WETH (a v4 pool with a custom hook only when the owner has allowed that hook). Before a buy it runs a buy-and-sell-back dry run, so a token that taxes or can't be sold shows up first. Returns and the day's result are counted in ETH, so ETH's own price moves are neither wins nor losses; dollar figures use the current ETH price. There's no $ARCIRCLE burn on this desk. Until its contract is set and funded it runs on paper and the page says so.

1. She finds every new Argus launch from Uniswap's PoolManager on Arc.
2. She reads each one: buys, sells and price per minute from the chain, liquidity and market cap from Dexscreener, and a full Token Scanner v3 scan.
3. Hard safety gates, which learning can never loosen: no critical flag from the scanner, launched at least 2 minutes and at most 3 days ago, at least $800 of liquidity, a buy and an immediate sell at trade size lose no more than 15%, taxes no more than 12%. The Token Scanner score is shown and learned from, but it doesn't block a buy: an Argus launch can't block selling.
4. Seven playbooks decide when to enter: Launch momentum (early buyers keep coming and it hasn't run away), Pullback (it fell 15–45% off its high and buyers are back), Volume breakout (volume jumped against liquidity and the price is breaking up), Steady climber (a calm, well spread token slowly climbing), Dex paid (the team just paid for its Dexscreener profile and the price hasn't run yet), Pump scalp (a sudden burst of buying or a fresh Dex payment: in fast, out at +20–25%) and Crash buy + DCA (down 55%+ from its high but still trading: $2, again at -15% and -30%, sell into the bounce).
5. She never trades $ARCIRCLE, and by default skips coins launched through ArcPad's own Argus flow, because the platform earns fees on those.
6. Real money has stricter rules than paper. An Argus pool starts at its launch price, and if the early buyers all sell, the price falls back there in one block. So a real buy needs the price at most 6× its launch floor (the pump scalp: 7×), a top-10 sell-off that would drop it less than 80% (the pump scalp: 60%), and no critical flag in the last 6 hours.
7. Then Claude gives a second opinion on the live numbers. It can only say no: it can never make the desk buy something the rules refused.
8. She also reads each launch's website, X and Telegram (from Argus's launch data and Dexscreener) and whether the team paid for a Dexscreener profile. A "Dex paid" playbook looks for entries after the payment, before the price runs. A link that another launch also uses is a copy-paste warning: no real money.

Every setup that passes the gates is also traded on paper (no money), so she learns from many more trades than a small wallet can pay for. Paper trades count half.

Each closed trade teaches three things: which playbook works (each playbook's average return, with the choice made by Thompson sampling), which setups work (a model over about 20 signals such as buy pressure, momentum, liquidity, holder concentration and the scanner's sections), and where to exit (every trade records when it first crossed +10% to +100% and −5% to −30%, and keeps being watched after it closes, so every take-profit and stop-loss pair can be replayed; once a day the exits move one step toward the pair that did best, and the stop-loss never goes wider than 25%).

Adding to a position (a second buy on a dip, or on strength) is learned before it's used: every paper trade records what one add would have done, and a kind of add goes live for real money only after 20+ cases with a clearly positive edge — never during the warm-up.

The first 25 real trades are a warm-up: every setup that passes the gates is taken, small, to learn fast. After that she keeps trying something new on a share of setups that shrinks to 12%, and picks the rest by what has worked. Each day she writes a journal on this page.

Profit is taken in steps, because the few launches that run 5–20× pay for all the losers: at the playbook's take-profit (about +30%) she sells 35%, at a double she sells another 25% (the stake is back by then), and the last 40% rides with a trailing stop 30% below its peak (40% once it's 5× or more), for up to 48 hours. After the first take-profit she never lets the rest fall below the entry. Every trade also has a stop-loss and a time limit. Between the once-a-minute checks she re-quotes what she holds every 10 seconds, so a stop-loss or a crash is acted on in seconds, not a minute later. After a real trade loses 30% or more, Claude writes a short review of what went wrong. She sells everything at once if the scanner finds a new critical flag, the price halves within a minute, or the pool can't be quoted anymore. Prices come from an exact quote from the desk contract, so pool fees, hook taxes and price impact are already in every return shown.

• A buy starts at 6% of the desk (at least $3) and is sized by risk: if the coin fell back to its launch floor, it would lose at most 1.25% of the desk. Size halves after 3 real losses in a row, when the last 20 real trades lost money in dollars, or once the day is down 5%, and drops to the $3 minimum when those trades lost twice what they made. A pump scalp that falls 5% within 2 minutes of the buy is sold at once. The desk's owner can tune these within fixed bounds by signing with the owner wallet; the settings can't make her trade.
• At most 10 positions open and 8 buys an hour; $2 always stays in cash; the contract's own limits cap each buy and each day. During the warm-up (the first 25 real trades): $3 a trade and at most 6 buys an hour. Each DCA tranche is $2. Since 30 Sep 2026 real money goes only through the Pump scalp; the other playbooks keep trading on paper so the desk keeps learning from them.
• After losing 15% in a day, no new real trades until the next UTC day (paper trading continues).
• Under a $5k market cap, a position is never held longer than 30 minutes; a crash-buy DCA position under a $10k market cap is sold after 1 hour.
• Once a day, 20% of new profit above the desk's previous high buys $ARCIRCLE and sends it to 0x…dEaD through the contract. Deposits are not counted as profit. The share may change later — not decided.

New coins are the riskiest thing on-chain and most of them lose value. ARCIA will lose trades, and early results are learning, not a track record. Nobody can deposit into the desk or copy it, and no message, chat or command can make it trade. Nothing on this page is financial advice.