The proof
Everything we claim, and how to check it
This page is for the sceptical reader. It has no urgency in it and nothing is rounded in our favour: what we measured, what the numbers were, what they do not prove, and where our own results were worse than we would have liked. Every whale named here has a public profile with its own receipts.
What we measured
We studied 1,300+ Hyperliquid wallets for 90 days. Following the most profitable ones blindly doesn’t work.
Not an opinion, and not somebody else’s study — this is our own measurement, and it is the reason the risk system exists.
78.4%
The wallet our own scoring ranked first looked superb
Across the 51 positions our selection grades it on, it won 78.4% of the time — the highest in the set on the number our own scoring ranked by. Any list built on measured win rate puts it first.
44.3%
On the trades a follower could actually take, it won 44.3%
We re-ran it across all 149 positions a follower could have entered, not just the ones our metrics grade. Worse than a coin toss. Same wallet, same 90 days — the difference is entirely which trades you count.
−0.06
Even our own expected-return model barely predicts what a copier captures
We measured the correlation between our own expected-return model — already risk-adjusted, our most careful attempt — and what a copier actually earns per position, across 130 wallets. It is −0.06: statistically nothing. If our own risk-adjusted ranking barely predicts it, raw profit certainly doesn't.
The same measurement indicted us
Across the five wallets we studied, our selection graded them at 68.0% over 247 positions. On everything a follower could have entered — 468 positions — it was 56.0%. What we actually delivered, over 264 closed signals in our receipt log, was 36.7%.
Twelve points of that gap came from grading a slice of trades nobody receives. The other nineteen came from a fixed stop rule of ours that closed 64% of positions early. We retired the rule and rebuilt selection around what a follower can actually capture. The replacement has closed too few signals to quote a number for, so we don’t quote one — check the whales’ own trade logs and watch it accumulate.
So what does predict a whale worth following?
Not profit. In our data a wallet winning 52% of its trades had a worst case 8.7× tighter than one winning 78%. The worst case is what a stop on that whale has to allow for: its deepest drawdown on one position or its widest stop, whichever is larger, plus one point of headroom. At identical risk, a worst case 8.7× tighter buys 8.7× the position. What predicts a whale worth following is how much of its own capital it puts at risk to earn what it earns — and that is what the colours publish.
Full method, sample sizes and the definitions behind every number: how we measured it.
How selection works
Six filters, measured on whole capital
These are admission thresholds: a wallet clears all six before it is listed, and every number below is the one the engine runs, read from its own configuration. After admission only the median hold and the account value are checked again, so a listed whale’s numbers today can sit below the bar it cleared — its page shows them as they are. A few whales on the board were hand-picked by us rather than admitted by these filters. The full method, with sample sizes.
- 1
At least 50 measured positions
We only rate a wallet once we have enough of its history to mean something.
Under 50, luck and a real edge look identical.
- 2
Win rate 60% or better
Measured after fees and funding, on positions a follower could actually have taken.
Win rate is the one trait we measured that genuinely persists between periods.
- 3
Max concurrent drawdown under 50% of whole capital
The deepest drawdown its positions put on the account while open together, over the positions we measured — against the entire account, spot and perps together, not one position in isolation.
Whole capital is the only denominator that can't be gamed by opening a second account.
- 4
Median hold of 2 hours or more
A copyability rule, not a quality one.
One wallet we tested ranked #2 of 20 on our quality composite and closed 71.3% of its positions before a follower could act. Excellent, and un-followable.
- 5
All-time profit is positive
A sign test, never a threshold — the account is up, or it is out.
Deliberately blunt: the size of the profit is what the other four filters argue about.
- 6
Account value of $100,000 or more
Measured on whole capital — every dex plus spot — and re-checked every night. A whale that falls below it leaves the board and says so on its page.
Two wallets on the board held almost nothing and a few thousand in profit, which made every ratio on their cards meaningless. A return on nothing is not a return.
Then: checked every night
Passing once is not a permanent seat. Every night each whale’s median hold and account value are measured again; a whale that fails either 3 days running leaves the board, and returns after as many clean days in a row.
Nobody is deleted — a whale that leaves keeps its page and its receipts, and can come back.
Risk, published
Every signal wears its risk
Colors come from the whale’s max drawdown per position on whole capital. The thresholds are public — the same math for every whale, no exceptions.
- worst single-position drawdown in the last 90 days is under 5% of account< 5%
- worst single-position drawdown in the last 90 days is 5–10% of account5–10%
- worst single-position drawdown in the last 90 days is 10–15% of account10–15%
- worst single-position drawdown in the last 90 days is 15–20% of account15–20%
- worst single-position drawdown in the last 90 days is 20% or more of account> 20%
- not enough verified history to rate—Shown on the site, never in our top list.
The rules we run
Every number here is read from the code that enforces it at the moment you load this page. If one of them changes tonight, this page changes with it.
How a package is built
- Who can hold a slot
- a whale that passes the board's gates, made a net profit over its last 30 days, and closed at least 20 positions in them
- How they are ranked
- by 30-day win rate — the card's own figure
- How many
- the highest-ranked whales whose signals — every entry they broadcast — fit 5 a day between them over their last 30 days; the ceiling decides, even below 3. A package sends only its whales' entries at its floor (below), fewer than they broadcast
- When a package ships short
- fewer than 3 whales qualify — the card says the real count
- White
- draws from white · default 1% per trade · worst member inside 5%
- Green
- draws from white · green · default 2% per trade · worst member inside 10%
- Yellow
- draws from white · green · yellow · default 2% per trade · worst member inside 15%
- Orange
- draws from white · green · yellow · orange · default 3% per trade · worst member inside 20%
- Red
- draws from white · green · yellow · orange · red · default 3% per trade · no drawdown bound
How membership changes
- Voluntary rotation, per package
- at most 2 in any 7 days
- A whale that stops closing trades
- leaves its packages after 21 days; it stays on the board
- A paused whale returns
- after 3 consecutive clean days
- What takes one off the board
- account value under the minimum · median hold under the minimum
- What takes one out of a package only
- a colour the package does not hold · too far below its 30-day high on closed trades · no closed trade in too long
- Minimum the whale itself commits
- 5% of its own account
- Minimum median hold
- 2 hours
- Minimum account to stay on the board
- $100,000
Receipts, not projections
Every number on this site is a closed position
Nothing here is a projection, a backtest of our own signals, or a simulation. A number appears on this site only after a whale opened a position and closed it, and we recorded both.
The log starts 2026-08-10. Everything before that carries numbers from a measurement bug we found and fixed, so we publish none of it — the feed is thinner than it would otherwise be, and that is the trade we chose.
Peak and realized are always separate. A peak is what a position reached at its best moment; realized is what it closed at. Quoting one as the other is the commonest way a track record gets inflated, and we treat it as a bug rather than a style choice.
Every figure is the whale’s. It is what they realized on their own capital. We do not know your position size or when you acted, so we never state what you would have made.
What we do not know
- Whether any of this repeats. Our measurement window is 90 days and one market regime. We have no walk-forward test across regimes, and every threshold we publish was derived on that window.
- What you will make. That depends on your size, your timing and your exchange. We report what other traders did.
- How our current risk model performs. We retired the rule that caused most of our delivery losses, and the replacement has closed too few signals to quote a number for. We would rather say that than quote one.
The full method, with every sample size and definition, is in our methodology note.
207 whales pass these filters right now. Their entries reach you the moment they open.
DarkTrade.AI reports what other traders did. Not financial advice.