Docs

How Rake actually works.

Rake runs a swarm of automated agents that quote both sides of prediction markets. Agents that make money get cloned. Agents that lose get killed. Everything on this page is measured, and where a number is modelled it says so.

The income

Polymarket pays a fixed pool every day to market makers who rest orders near the middle of the book. The score is ((max_spread - our_spread) / max_spread)^2 weighted by order size, sampled once a minute at a random offset. At 00:00 UTC the day's pool is split between makers by their share of that score. Which way the market resolves never enters into it.

The cost

A resting order gets taken precisely when the price is moving against it. That is adverse selection and it is the whole game. In our backtest over 30 days of real history it took back more than twice what the rewards paid, until the strategy was changed. It now takes 23% of them.

What we measured

-36.34Naive quoting, $/day
+79.37Tuned, pairs merged, $/day
+65.04Held-out half, $/day
52 / 60Markets that qualify
30Days of history

Backtest on real Polymarket price history, 60 markets, no money at risk. Reward share capped at 33% per the limits published by the leading open-source maker. Not a prediction of future returns.

The three rules that matter

Rest the minimum. An order only scores if it is at or above the market's minimum reward size and inside the maximum spread. Anything above the minimum just enlarges the position you are left holding when someone gaps the book through your quote.

Skip the busy books. Filtering out markets whose hourly price moves exceed 2.5 cents of realised volatility is the single change that moved the backtest from losing to positive.

Merge the pairs. Both legs are buys: yes at a discount, no at a discount. One of each is worth a dollar at resolution whatever happens, so a matched pair merges back to USDC and banks the difference. Counting that merge, rather than marking both legs as open positions, is what the 30-day figure above rests on.

Honest limits

The edge is concentrated: five markets produced 55% of the modelled profit, and 45 of the 52 made money. Out of sample it comes in about 14% lighter. The token is not launched and there is no contract. Figures on the home page are modelled. Market making can lose money and regularly does.

Verify

Telegram t.me/rakemarkets · domain rake.markets · desk wallet 0x7b3a…c4e2 · token not launched. Anything else claiming to be Rake is not us.