ArcScore rates tokens on the Arc blockchain from on-chain data — read by us, apart from the few third-party readings listed under "Still external". Every token gets seven sub-scores between 0 and 100, combined into one total by fixed weights. No manual adjustments, no paid listings, no opinions. The same rules apply to every token.
Candidates come from the arcts token list, new Uniswap V3/V4 USDC pools found on-chain, GeckoTerminal pool lists (pointers only — every pool is then read by us), on-demand scans and every token we rated in the last 48 h. A token is rated only when it passes ONE admission gate that we measure on-chain ourselves, the same for every token: G1 at least $1,000 of real USDC across all its pools (live positions, a quoter sale through the pool's hook, or the pool's USDC balance) and G2 at least 20 wallets holding ≥ $10. New pools are checked by an hourly candidate job; a rated token is re-rated every run, its gate verdict is re-checked at least hourly, and it leaves only after failing the gate for 24 h. A venue we cannot read yet is flagged as unverified — never counted as empty. New tokens carry a NEW tag for 24 h; a confirmed rug stays visible as liquidity pulled.
Read from the Arc chain by us: liquidity and price, supply and burn, contract and pool creation, the admission gate (real USDC in every pool, wallets holding ≥ $10), sell checks (a quoter round trip on Uniswap V3 and V4 pools, hooks included, and a transfer simulation from a real holder), every swap in each rated token's main pool and its V4 pools — V2, V3 and V4 events (24 h volume, number of trades, unique wallets, buyer flows, round-trip share, 24 h price change; extra V3 fee-tier pools count toward liquidity, not yet toward volume), every token transfer (holder counts, wallet clusters) for tokens whose full history is indexed, new pools, USDC minted and burned on Arc, chain activity. Still external: holder counts from arc-scan for tokens whose history is not yet fully indexed, the explorer's lookalike / reserved-name flag, GeckoTerminal pool lists (candidate pointers only: every pool is read by us, no GeckoTerminal number is scored), and the arcts token list as one of three universe sources (names, logos). No volume, liquidity, trade count, market cap or price change from GeckoTerminal or arcts is published or scored for a rated token: when we have not read a number, it is shown as n/a. The one remaining arcts number is the liquidity of a same-symbol copy we have not measured yet (lookalike check), labelled on the card. Every figure on a token page carries its source.
Since 30 Oct 2026: liquidity = the real USDC a seller can reach, summed over every pool of the token. Every pool on Arc is found from the chain itself — the pool-creation event of every DEX factory (Uniswap V2/V3, Aerodrome and others) and every V4 pool initialised in the PoolManager — so a pool cannot be missed, new ones included. Each pool is read on-chain by us every run: a V2 pair its USDC reserve, a V3/concentrated pool its USDC balance, a V4 pool its live positions' USDC side (or, where those cannot be rebuilt, what our quoter shows a sale can actually get out; a pool where every sale reverts reads $0). No doubling, and the token side is never counted as liquidity — a pool holding only the project's own tokens is not a market. The both-sides value (USDC + token side at our price, comparable with DexScreener) is shown next to it as TVL and is never scored. A rug is a claim with evidence: a drop in real USDC is called a rug only when the removal is traced on-chain to the project (deployer, owner or pool creator) or the price collapses with it; officially bridged tokens (e.g. Chainlink CCIP cross-chain tokens, redeemable on their home chain) are recognised automatically and never called a rug.
Before 30 Oct 2026 (archive): A V2/V3 pool: its USDC balance × 2 (both sides). A V4 pool: its live positions (USDC side + token side at our price) when they reconcile with the pool's live liquidity; otherwise the pool reserves, but only when our quoter sell ladder (simulated sales of 0.1–50 % of the pooled tokens) confirms them; otherwise what that ladder proves a sale can extract, × 2. A cirBTC/WETH/EURC-quoted pool is valued with our own on-chain price of that asset. Where no reserves can be read, a quoter depth estimate is used and named on the card; a reading above 25 × the market cap (and over $500k) is cut to 2 × market cap. The admission gate (G1) uses the USDC side alone — the "real USDC" of the token. Market cap = the price shown × on-chain circulating supply (total supply minus burn addresses).
Liquidity (25%) — pool depth (above) and how the liquidity is controlled (launchpad-managed vs. a manually created pool whose owner can withdraw it).
Concentration (15%) — how much of the supply sits in the 10 largest wallets, after removing the pool, lockers, burn addresses and contracts, and whether one wallet holds over 25 %; for tokens whose full transfer history we index, a large linked-wallet cluster, deployer-linked wallets holding ≥ 10 % and top wallets funded at launch lower it further.
Holder quality (15%) — holders holding ≥ $1, the committed core (wallets ≥ $100 and ≥ $1,000), the dust share of buys and the 24 h buyer/seller balance: distinct wallets that bought / sold at least $1 in our own swap index (wallet = transaction sender), weighted by dollars so a swarm of tiny swaps counts for almost nothing; with 1–4 such wallets the balance is blended toward neutral, and a token whose main pool we do not index scores it neutral.
Lookalike risk (10%) — explorer name checks and same-symbol tokens on Arc. The original is the token people actually hold: at least 50 holders and at least 3× the holders of the largest known copy (when no copy has a known holder count: the most real USDC, at least 5× all copies combined — a copy we have not measured ourselves is compared by arcts' liquidity figure, and the card says so). The original is only lightly marked for buyer-confusion risk, because being copied is not its flaw; copies score lower.
History (15%) — age, deployment path (launchpad, factory, direct deploy) and whether the source is verified.
Trading pattern (10%) — volume relative to liquidity (an idle pool under 5 % daily turnover does not get full marks), buy/sell balance and how many distinct wallets stand behind the trades. Round-trip (wash-like) volume — the share of volume from wallets that both bought and sold within the same day, the worse of the last 24 h and the 7-day daily mean — lowers this score continuously from 50% upward once there are enough traders or swaps (from 30 traders / 100 swaps, in full at 45 / 150); a net USDC outflow above 40 % of the 24 h flow (on at least $5k) lowers it too.
Volatility (10%) — drawdown from the 24 h high and the size of the 24 h move. The 24 h change is our own: the volume-weighted price of our indexed swaps (last 3 h vs 21–27 h ago, at least 20 swaps in each window) and our on-chain spot price archive (now vs 24 h ago); when both exist, the less extreme of the two is used, and the card shows both. With neither, a VWAP from 5 swaps per window is used and labelled "few swaps".
Unknown is never perfect. An input we could not read is scored neutral and flagged, never as the best case: an unknown 24 h change scores volatility 60 at most, an unknown dust share 60, an unknown holder list 50, a partial holder count is a lower bound (no small-holder cap). A missing core input (holder count or holder list, 24 h volume, our own liquidity reading, or V4 hook-pool liquidity the quoter cannot confirm) makes the token NOT RATED: the facts are shown, no number is published.
One rule for every token: G1 at least $1,000 of real USDC across its pools and G2 at least 20 wallets holding ≥ $10, both measured on-chain by us. A rated token that starts failing the gate keeps its listing for 24 h and its card says so — failing the entry rules since …, removed if not fixed by …; a pass in between resets the clock. A venue we cannot read is flagged as unverified, never counted as empty: when G2 passes and G1 cannot be decided only because a venue or a hook reading cannot be confirmed by us, the token is listed with a visible "unverified" note on its card instead of being dropped.
The total is the weighted average above, then the LOWEST of these caps applies; the card names the cap that decided (score capped at N: reason). No cap switches on in one step: each measured cap starts at 100 (no cap) at its threshold and falls linearly to its full value, so a small change in the data moves the score a little, never 20 points at once. Where a rule has a condition (enough traders, enough holders, a pool size), the condition fades in over the stated range too. Facts — a rug, a blocked transfer, a reverting sell, a copy — apply in full.
Wrapped/bridged assets are exempt from the round-trip, thin-market, farmed-holder and lifetime caps (their round trips are arbitrage, their pools a parking place). Informational flags such as LP unlocked, one wallet holds N% or N same-symbol tokens explain what pulled a score down.
A ≥ 80 low risk · B 65–79 moderate · C 50–64 elevated · D 30–49 high · F < 30 critical. A letter changes only when the total is at least 2 points past a line for 2 refreshes in a row (a jump of two letters is accepted at once), so a token sitting on a line does not flicker.
The Arc chain itself — historical event logs through HyperSync, new blocks through a websocket stream, state reads through public JSON-RPC nodes — decoded by us into our own swap/transfer index and every reading above; arc-scan.org (holder counts for not fully indexed tokens, explorer flags), GeckoTerminal (candidate pool lists and pool pointers only, never a scored number) and arcts.fun (token list, names, logos). Blockscout is used only as an independent second method in our proof checks; no published number comes from it. Scores refresh every few minutes — a full pass of all tokens takes several minutes, and the exact age is always shown on the page as "X min old"; slower-moving readings are cached — holder counts, top holders and supply up to 1 h, the sell check up to 6 h — and a token's creation block and deployer, which never change, are read once. Every token page shows the exact on-chain evidence behind each sub-score.
LP lock status is inferred from how the pool was created, not from reading every LP position. The sell check is a simulation at the current block; a contract can change behaviour later. Scores describe risk structure, not expected returns. Nothing here is investment advice.
Machine-readable scores: data.json.
We sum the USDC leg of every swap from the V2, V3 and V4 pool events on Arc, swap by swap, from our own index. Other trackers report far larger 24 h volumes for many Arc V4 pools — in our checks 3–25× ours. The trade counts match ours one to one and the USD value of each individual trade matches to the cent; only their totals do not. We therefore show the sum of real swaps, and next to it the round-trip share (the part of that volume made by wallets that both bought and sold within 24 h), which no total can show. When a token trades mainly on a venue we do not index yet, we say so on the card: its volume is n/a and the token is not rated until we read that venue. We never fall back to a tracker's figure.