A memecoin health check. Paste a Solana or EVM token address (or a ticker) and token-risk reads the contract, the pools and the holder list, then returns a 0-100 grade with every finding behind it.
Live: https://token-risk.vercel.app
| Vital | Weight | Signals |
|---|---|---|
| Contract & authorities | 30% | Solana: mint / freeze authority, Token-2022 permanent delegate and transfer fees, mutable metadata. EVM: honeypot simulation, buy/sell tax, mintable, reclaimable or hidden owner, unverified source, proxy, pause, blacklist |
| Liquidity | 25% | Pooled USD (log-scaled), liquidity-to-market-cap ratio, share of LP locked or burned, CEX listings |
| Holders | 20% | Top-10 and top-1 share with AMM pools, lockers and burn addresses removed, deployer-linked wallets, creator balance, holder count. Lorenz curve, Gini and HHI on the top 20 |
| Trade flow | 15% | Buy/sell ratio, volume-to-liquidity turnover (dead market vs wash trading), 1h crash and 24h volatility |
| Age & presence | 10% | Oldest pool age, listed website and socials, paid DexScreener boosts |
Disqualifying findings cap the total no matter how good the rest looks: live freeze authority caps at 35, live mint authority at 50, honeypot at 5, liquidity under $10K at 40.
Exit test: price impact of selling $1K to $100K into the pooled liquidity using the constant-product formula impact = x / (R + x) where R is the token side of the pool in USD, plus the largest sale that stays under 2% and 10% impact.
Tested on pump.fun coins at the moment they graduate to a DEX pool (live results, code in validation/).
- Data: every coin pump.fun lists as graduated (about a day's worth). For each one, the state at graduation is rebuilt from one-minute candles, metadata and the creator's earlier launches, then labelled by what happened next. The outcome is "price 6 hours later is 90% or more below the graduation price".
- Result: the 0-100 score above cannot separate pump.fun graduates (AUC about 0.5), because at that moment they all look alike. A logistic model on launch behaviour reaches AUC about 0.87 under 5-fold cross-validation grouped by creator. Time to graduate is the strongest signal: 84% of coins that graduated within their first minute were down 90% six hours later, against 24% of those that took over 30 minutes.
- Mayhem Mode: about a fifth of "graduated" coins are Mayhem coins whose curve completed with a fraction of a SOL, after the agent sold its extra 1B tokens into it. They never reached the graduation price and are reported separately.
- Live test: a GitHub Actions job (
.github/workflows/collect.yml) records new graduations every 30 minutes with the full score and the model's prediction, then labels them after 1h, 6h, 24h and 7d. Results go to thedatabranch.
For pump.fun coins the scan adds a launch section: Mayhem, Holder Rewards and Cashback notes, phantom-graduation detection, and the launch model's probability with its main drivers.
node validation/build-retro.mjs # historical dataset (about 50 minutes, rate limited)
node validation/analyze.mjs --fit # cross-validate and refit lib/launch-model.json
node validation/collect.mjs # one forward snapshot and labelling passDexScreener (pairs, liquidity, flow, boosts) · RugCheck (Solana authorities, holders, LP locks, insider graph) · GoPlus (EVM contract simulation and holders) · GeckoTerminal (hourly candles) · pump.fun (launch data, candles; unofficial endpoints)
api/scan.js resolve address or ticker, gather sources, score
api/trending.js tokens currently buying DexScreener boosts
lib/sources.js data adapters and normalisation
lib/score.js the scoring model (pure, shared with the browser)
lib/launch.js pump.fun launch features, labels and the logistic model
lib/pump.js pump.fun client (throttled)
validation/ dataset builders, analysis, forward collector
test/ node:test unit tests
index.html report UI, vanilla JS
npm test
npx vercel devA screening tool, not financial advice. A clean report does not stop a team from selling.