How a case file is built
Sherlock doesn't guess. Every answer is written from exhibits you can open, and every exhibit comes from public data: Robinhood Chain, X and market history. Here is what each one reads and where it falls short.
Where the evidence comes from
- Robinhood Chain (chain ID 4663): an indexer reads every block from genesis into a database of tokens, trades, balances, liquidity positions and native ETH transfers between wallets. Live reads go through the public RPC.
- DexScreener: price, market cap, liquidity, volume and pool age for the token you paste, the moment you paste it.
- X: public posts that mention a ticker or contract, with their engagement captured over time.
Sample tokens in the app run on a simulated version of this data so the whole product can be tried today. Real tokens show live market data only until the indexer and X sources are connected; Sherlock never attaches simulated findings to a real token.
A · Bundles
A bundle is a set of wallets controlled by one party that buy at launch so the supply looks distributed. Sherlock takes every wallet whose first buy landed in the first blocks after liquidity was added, then walks the funding graph backwards: who sent each wallet its first ETH, when, and how much.
Wallets are grouped when they share several of these signals:
- The same funding address, directly or one hop back.
- Funded inside a short window before launch, usually minutes.
- Identical or near-identical ETH amounts.
- First buys in the launch block or the next few, often with matching gas prices or router paths.
Each cluster shows its signals and a confidence level. The panel reports what the cluster bought, what it still holds, and that bag's size against the pool's liquidity, because a bag bigger than the pool is what hurts people when it sells.
B · Deployer
The wallet that created the token contract, where its first ETH came from, and every token contract it created before. Each earlier launch is marked as still trading, abandoned, or rugged (liquidity pulled), with how long it lasted. A deployer whose earlier launches mostly died within 48 hours is the strongest single signal in a case file.
C · Origin & lineage
Most launches start as a post. Sherlock looks for the earliest post that names the ticker, the contract or the idea before the token was deployed, and measures its engagement against other origin posts from the last 30 days. Lineage links a token to earlier tokens on the same meme: same origin post, same name, same art. A derivative isn't a red flag on its own, but it shares attention with every other copy.
D · Comparables
When you ask what a token could reach, Sherlock answers with what similar tokens did: launches from the last 30 days with a similar narrative and origin engagement. It shows their peak market caps (the middle half as a range, and the median), how long they took to peak, and how far they've fallen since.
This is context, not a forecast. The drawdown sits next to the range on purpose: most comparables are far below their peak.
E · Holders
Balances for every holder, with the liquidity pool separated out. The panel shows the top 10 share without the pool, which top holders belong to a bundle or the deployer, and how many holders are fresh wallets created within a day of their first buy, a common way to inflate a holder count.
F · Contract & liquidity
- Ownership: renounced, or an owner who can still call admin functions.
- Permissions: mint, blacklist, pause, and whether the contract is an upgradeable proxy.
- Tax: buy and sell tax read from the contract.
- Sell simulation: a fresh wallet buys and sells on a fork of the current block. If the sell reverts, the token is flagged as a possible honeypot.
- Liquidity: LP tokens burned, locked (with the unlock date) or free.
The risk read
The verdict is rule-based, not a model's opinion. Each flag adds a fixed number of points (a bundle holding more than 20% of supply, a deployer with a record of short-lived launches, unlocked liquidity, a failed sell simulation) and the total maps to low, moderate, elevated or high. Every point comes from a flag you can see in the panel. A low score says nothing about where the price goes.
The written answer
A language model reads the case file and your question and writes a short answer that leads with the direct reply. It's instructed to use only numbers from the evidence, say when something isn't in the case file, and never tell you to buy or sell. Your question and the case file are sent to the model provider for that request only. See the privacy policy.
What it can't see
- Wallets funded long before launch through an exchange look independent onchain, so some bundles are missed.
- Deals, promises and team wallets arranged offchain.
- Posts that were deleted before they were captured, and activity on platforms other than X.
- Anything that happens after the block the case file was read at. Every case shows that block.