How cream.run works

cream watches a small, measured group of the most profitable traders onchain and turns what they're actually holding into two things: a vault per market (long, short, or flat) and one fund that trades the highest-conviction signals as a single long/short book. This page explains the whole machine: who gets listened to, how a verdict is formed, and how the fund executes it.

1 · The big picture

Some traders are verifiably, consistently good. They've made millions over years, in public, onchain, where anyone can check. cream.run tracks a curated group of them. Every 5 minutes we take a snapshot of what each one is holding, combine those positions into a single verdict per market, and each vault mirrors that verdict. The Fund then allocates one portfolio across the strongest of those signals.

No prediction models, no black box. The only input is what proven traders are actually doing with their own money, right now.

2 · The roster: who gets listened to

The roster starts from Hyperliquid's public leaderboard, where every wallet's performance is visible to anyone. From there it's a funnel: hard filters throw out the noise, a composite ranking keeps the best 200, and then we measure everyone ourselves before a single vote counts.

Hard filters: a wallet qualifies with at least $50K of account value, at least $100K traded in the last 30 days, activity within the past week, and lifetime profitability. Qualification earns observation, not a vote: every 5 minutes we snapshot each wallet's positions and equity and score it the way a hedge fund scores its managers, building a skill score q from measured Sharpe, Sortino, and max drawdown, percentile ranked against the whole cohort and re-scored daily.
Ranking: qualifying wallets are ranked on a weighted composite of lifetime PnL (half the weight), 30-day return, and 30-day volume. The top 200 constitute the tracked roster, refreshed daily.
The voting bar: a wallet votes only after ~10 hours of observed history and only while its measured skill clears the quality bar. Fall below it, and the wallet keeps being watched but stops being heard.

3 · The voting room

Picture each market as a room. Every voter with a position in that market walks in and votes: long (price goes up) or short (price goes down). The vault holds whatever the room decides, but not every voice is equally loud. Two things set a trader's volume:

Track record

The main thing. The skill score q from the roster funnel: measured, not claimed. It counts disproportionately: someone twice as skilled speaks four times as loud.

Skin in the game

A trader betting their whole account believes it more than one dabbling with 5%, so bet size raises their voice, but only gently. A maxed-out bet makes you less than twice as loud. Skill first, bet size second: one reckless whale can never out-shout ten skilled traders.

4 · From vote to position

The room's verdict is a number between −100% and +100%: how lopsided the vote was. The vault simply holds that fraction of its book in the market and keeps the rest in cash. A verdict of "40% short" means 40% of the vault's book is short and 60% sits safely in dollars.

Positions are always taken at 1×, with no leverage. The position is never bigger than the money behind it, so a bad day is a bruise, not a wipeout, and liquidation is remote by construction.

The vault also never copies individual trades. It reacts to positions: a trader's trade only matters if it changed what they're holding by the next 5-minute snapshot.

5 · The Fund: one book, the strongest signals

The vaults page shows every market's verdict on its own. The Fund is what a portfolio built on those verdicts looks like: a single long/short book that sizes each signal by how much there is to trust it, caps everything so no single idea can sink the book, and re-checks itself every 5 minutes.

Each market's target weight is simply signal × confidence. The signal decides the direction; confidence decides the size. Confidence multiplies three measurements, each answering a different question:

Breadth

How many independent voices agree? An effective count of real voices, so a verdict carried by one loud wallet sizes small even when it's extreme.

Skill

How good are the specific traders holding this market right now? The average measured skill score of the actual holders, not the roster in general: the people behind this call.

Edge

Has this market's signal actually been working? Recent risk-adjusted performance of the signal's own track record. A signal that's been losing sizes toward zero until it earns trust back.

Conviction floor: targets below 10% of the book are skipped entirely. A bet too weak to size is a bet not worth paying costs on.
Per-market cap: no single market can exceed 50% of the book, however convinced the room is.
No leverage: all bets together are capped at 100% of the book; if raw targets ask for more, everything scales down proportionally. Capital not deployed behind a signal stays in USDC; that's the "idle cash" row in the allocation table.
Execution: the book re-checks its signals every few minutes and trades only when they meaningfully change. The exact trading rules are part of the fund; the fills in the order history are their public record.

6 · Reading the site

The vaults table is one row per market:

Days: how long this vault's live track record has been running.
Call: the room's current verdict: Long, Short, or Flat (in cash).
Allocation: how much of the vault's book is deployed in the position; the rest is cash.
Pros: how many tracked traders currently hold a position in this market.
Vs Hold: the vault's return compared to simply buying and holding the asset over the same period.

The Fund page is the portfolio view of the same machine:

Fills tape: the scrolling strip up top: every recent paper fill as an arrow (▲ long, ▼ short) with its execution price.
NAV chart: the fund's value, indexed to 100 at inception, with the live point pulsing at the edge. The long/short bar above shows how the book splits.
Allocation: every current bet as a share of a 100% budget, its side, and its PnL (that bet's contribution to the fund's value since inception). The USDC row is the idle cash from the rails above.
Latest signals: the decisions as they happen: Open, Add, Trim, Close, Flip, each with the resulting size.
Order history: the raw fills behind those decisions: side, time, and the mark price each rebalance executed at.

Each vault page adds four views:

Price chart: the asset's real market price (from Lighter), with the vault's position drawn over it: green above the midline when long, red below when short, flat at its latest value.
Positioning over time: the raw verdict history: how the room's lean has moved, tick by tick, since day one.
Vault vs. holding: two lines starting from the same point: the vault's value and a simple buy-and-hold. The gap (Diff) is what following the smart money added or cost, over 1 day, 7 days, or all time.
Who's driving it: the actual traders behind the current call: their direction, how much of their own book they've committed, and their skill score.