RAKE

How RAKE works

The problem, in one story

You find a token on Base. It's pumping: $1.1M volume, +35%, 4,000 swaps. Looks exciting.

But volume doesn't tell you the story that matters. Suppose the truth underneath is:

$1.1M entered the pool. $506K left through wallets that were there in the first block, fund each other, or sell every single window. One repeat seller alone took $178K.

Now your question changes from "is this pumping?" to the only question that matters:

"How much of this market is real, and how much is money being extracted by people who were already positioned - am I trading against the house?"

Dexscreener tells you what happened to the price. Basescan tells you which transactions happened. RAKE tells you who got paid.

What actually happens when you pull the tape

  1. Find the pool. RAKE resolves the token's top-volume pool on Base (Uniswap v2/v3/v4, Aerodrome, PancakeSwap - v4 pools live inside a singleton contract and are handled specially).
  2. Reconstruct the tape. Every Swap event in your window is read from Base logs and decoded.
  3. Attribute the humans. Each swap is billed to the person behind it. Smart-wallet transactions (ERC-4337) are resolved to the actual account, never the bundler that submitted them; routers and aggregators are never blamed for their users' trades.
  4. Price honestly. Every swap's USD value comes from its own WETH/USDC leg at the hour it executed - not from a spot ticker after the fact.
  5. Classify sellers by rules. See the cohorts below. Rules, not opinions.
  6. Walk the money. With funding data, RAKE checks who first funded the biggest sellers - catching fleets of "independent" wallets that share one low-degree funder.
  7. Compute the rake. The share of the window's inflow that left through house cohorts.
  8. Let the analyst read it. An LLM writes the diagnosis from the finished evidence. It may spend a small budget of extra funding walks; it cannot produce or change a number.

The house cohorts

CohortRule (mechanical, reproducible)
first-blocktraded in the first 50 swaps after the pool was created - the snipers
deployer-fundedfirst inbound transfer ever came from the initial-LP wallet or a first-block wallet
clustersold alongside other sellers who share the same low-degree first funder - one operator, many hands
lpadded or removed liquidity in the window, plus the pool itself
repeatalso sold in the previous window of equal length - sells every window, every time
unlabeledeveryone else

Important honesty note: "house" does not mean "proven to be the same person" or "the dev". It means: wallets exhibiting behaviors that fit predefined extraction rules, with the evidence attached. Exchange hot wallets, bridges, and wallet-infrastructure funders are explicitly excluded from clustering so ordinary users are never smeared. If a judge asks "how do you know they're the same entity?" - the answer is: we don't claim identity; we identify economically connected behavior using reproducible rules, and the receipt exposes the evidence.

Who uses this, and when

  • Traders, right before buying. Paste the token. If 46% of today's inflow left through house cohorts, you may still buy - but with $200 instead of $1,000, and with open eyes.
  • Traders, right after getting wrecked. Add your wallet and get your ticket: who sold, in size, within seconds of your entry. "Who sold to me?" finally has receipts.
  • Token communities and founders. When X says "DEV IS DUMPING", stop arguing and publish the receipt - whichever way it comes out. Watch your own token with the Telegram guard and know about coordinated extraction before the community does.
  • Researchers and investigators. Paste token, get investigation: every figure traces to a transaction, so the receipt is quotable.
  • Trading bots and AI agents. Call the deep pass, pay a nickel, get structured forensics before deciding to trade. RAKE is infrastructure for other agents, not just a page.

The self-check

Every hour the agent rakes Base's trending tokens and appends the events to a public GitHub log. Twelve hours later it checks each event: did price actually fall after a high-rake window? The split - high-rake vs baseline - is published even if the answer is embarrassing. A signal you can't audit is an opinion; RAKE's log is the audit.

What RAKE is not

  • Not a bubble map. Bubblemaps shows you the ownership network; RAKE quantifies the extraction. Different question, different tool - they compose well.
  • Not a buy/sell signal. RAKE gives evidence; you make the decision.
  • Not an oracle. Thin windows read TOO THIN; unreadable pools read UNPRICEABLE. It refuses to guess.