I Want to Start With Arithmetic, Because Arithmetic Is Where I Go When Other People Go to Hope
In the summer of 2020, auditing Compound Finance's governance module with a team of four developers spread across three time zones, I found myself at two in the morning staring at a reward distribution function that was beautiful in the way only a quietly wrong thing can be. The code was clean. The comments were warm. The manifesto — which I had read twice before opening the file — promised a system that treated every participant alike. The function did not. It treated whoever arrived first as though they had also arrived best, and it did so with a precision that made the unfairness look like engineering.
The most dangerous number in a protocol is the one that looks like a success.
I have carried that sentence for six years. It is why, when a screenshot crossed my feed in this February 2026 bull market showing an account sitting on roughly $5.78 million of unrealized profit — a 112% return on something described as a 25x Ethereum long and a 40x Bitcoin long — I did not feel the small warm flicker of envy I was apparently meant to feel. I felt the specific cold that arrives before an audit. Not because the number was fake. Because the number was true, and being true is not the same as meaning anything.
Here is what I saw in the eighteen seconds I spent looking at that image. I saw a market that has decided, collectively and without much discussion, that leverage is a personality. I saw a bull market doing what bull markets do — converting one person's private trading result into a public invitation. And I saw, scrolling beneath it, the thing that actually worries me: rows and rows of people asking which platform, which wallet, which entry, so that they could do it too.
The headline said a trader made money. The headline was accurate. It was also, in every way that matters to the people reading it, a lie of omission. What follows is the audit I would have performed if someone had asked me to verify any of it. I did not need their permission to do the math. That is the beautiful and terrible thing about on-chain finance: the arithmetic is public even when the disclosure is not.
The Instrument Nobody Describes Correctly
Let me build the machine before I turn it on.
A perpetual future is a contract with no expiry date that tracks an underlying asset through a mechanism called funding. You are not buying Bitcoin. You are buying an agreement about Bitcoin's price, settled in a currency the exchange defines, collateralized by a margin balance you post. The contract never expires, which means the only way a position ends is if you close it or if the exchange closes it for you. That second possibility is the entire story.
What makes the perpetual useful is also what makes it lethal. Because there is no settlement date, the contract's price has no natural gravity pulling it toward spot. So the protocol invents gravity: every few hours, one side of the book pays the other side a fee. If the perpetual trades above spot, longs pay shorts. If it trades below, shorts pay longs. This is the funding rate. It is a small number — often 0.01%, sometimes 0.1%, occasionally something that should make you sit down — and it is charged on notional, not on margin.
That distinction is where most retail readers lose the thread, so let me hold it up to the light. If you post $10,000 of margin and open a 20x position, your notional is $200,000. A funding rate of 0.01% charged every eight hours costs you $20 per interval on that $200,000, which is $60 a day, which is roughly $1,800 a month, which is 18% of your posted margin. The asset does not have to move against you for you to lose money. It only has to sit still while everyone else is also long.
In my 2022 research on Celestia's modular architecture — a 30,000-word analysis I published under the title "Sovereignty Through Separation," mostly because I needed to keep my hands busy during the worst of the bear — I spent a chapter on what I called settlement gravity. The point was simple: every system has a cost of existence, and in perpetuals that cost is charged to conviction. You are not paying for a position. You are renting the right to hold an opinion.
Now scale that. A notional book in the low nine figures pays rent in six figures per day when funding is crowded. I will come back to this, because it is where the 112% figure quietly dies.
The Plumbing Underneath: What Kind of Venue Is This
The report, as it reached me, declined to name the platform. That omission is not neutral. Where a leveraged position lives changes what its risk actually is, and the difference between venues is not cosmetic.
The current generation of high-leverage crypto trading clusters on perpetual decentralized exchanges, and one of them dominates the conversation: Hyperliquid. I have followed it closely since its HYPE distribution in late 2024, and I have written before, in these pages, about why its rise deserves both admiration and scrutiny. It runs its own chain, its own matching engine, and — this is the part that matters — its own price oracle, fed by a validator set rather than by a third-party data network.
That design is fast. It is also a governance surface. In March 2025, when a thin, low-float asset called JELLY generated a position that threatened the protocol's counterparty vault, the response was not algorithmic but political: validators voted to delist the market and settle positions at a price of their choosing. I do not consider that a scandal. I consider it a disclosure. When a system's stated invariants can be suspended by a vote, the invariant was never the invariant — the vote was.
This is the reason I have never fully bought the enthusiasm for modular data availability as a category. The celebrated separation of execution, settlement, and data availability layers is elegant on a diagram, and it answers a question — who stores the bytes — that the market has repeatedly demonstrated it will not pay much to have answered. The question the market actually pays to have answered is: who can stop my trade, and under what procedure. Data availability layering does not answer that. Governance does. And governance is not a layer. It is a group of people with names.
So when a whale's leveraged position is described without naming its venue, the description has stripped out the single most load-bearing fact. A 40x long on an unstoppable oracle and a 40x long on a vote-switchable oracle are not the same instrument. They do not carry the same tail. They do not deserve the same headline.
The Counterparty Is Not the Market. The Counterparty Is Someone.
Here is the sentence I want tattooed on the inside of every retail trader's skull: for every unit of floating profit, there is a unit of floating loss, and on a perpetual exchange the pool that eats the loss is often the community itself.
On a fully order-book-matched venue, the short side of that whale's long is held by other traders — market makers, arbitrageurs, hedgers, and the occasional person who read the same chart upside down. On a venue where a protocol vault acts as counterparty of last resort, the short side may be the vault, and the vault's capital is deposited by users seeking a passive yield. Those depositors are not anonymous institutions. They are the same retail crowd scrolling past the screenshot.
I first learned to read this structure during the Compound governance audit in 2020. We found a subtle flaw in a reward distribution algorithm that disproportionately favored early participants — the people who arrived when the incentive was small and the risk was large — and quietly disadvantaged the late joiners who supplied most of the liquidity. The protocol's manifesto promised egalitarianism. The code delivered seniority. I wrote 5,000 words about it under the title "The Hypocrisy of Decentralized Centralization," and it traveled further than anything I had written before, for the simple reason that people recognized their own experience in it. They had felt the shape of the flaw before anyone described it.
The shape here is different but the geometry rhymes. A headline about a whale's profit is, from the vault's perspective, a headline about a vault's drawdown. It reads as triumph on one side of the ledger and as expense on the other. Both readings are correct. Only one gets retweeted.
This is the part of the story I find genuinely melancholy. We built these systems so that no single party could quietly extract from the many. And in a bull market, the most reliable extraction mechanism turns out to be a screenshot.
Now Let Us Do the Arithmetic the Headline Refused To
I want to reconstruct the position from the two numbers given — $5.784 million of unrealized profit and a 112% return — because the reconstruction is where the story becomes interesting, and where it stops being credible.
Return on margin equals the price move multiplied by the effective leverage. That is the first identity, and it is unforgiving.
If the reported 112% is profit divided by posted margin, then the margin was approximately $5.784 million divided by 1.12, which is about $5.16 million. Fine. Now apply the stated leverage. At 25x, a 112% return on margin requires an adverse-to-you-but-favorable price move of 112% divided by 25, which is 4.48%. At 40x, it requires a move of only 2.80%.
Sit with those two numbers for a moment. The entire celebrated profit was manufactured by a price move smaller than an ordinary Tuesday. Bitcoin routinely travels three percent inside a single New York afternoon. Ethereum does it while you are in a meeting. The whale did not catch a wave. The whale caught a ripple and installed a lens in front of it.
Now invert the ripple. If a 2.8% favorable move produces $5.784 million on the Bitcoin leg, then a 2.8% unfavorable move produces negative $5.784 million on the same leg, and that leg is gone. Not reduced. Gone. At 40x with a maintenance margin requirement in the vicinity of half a percent on a tier-one asset, the liquidation price for a long sits roughly two percent below entry. Two percent. That is inside the noise band of a single funding cycle.
The source material I was working from suggested that a Bitcoin or Ethereum correction of 3.5% to 7.5% would be needed to trigger partial liquidation. That estimate does not reconcile with 25x and 40x, and the discrepancy is the most informative thing in the entire report.
There are only two ways to close the gap, and both are damning.
The first possibility is that the quoted 25x and 40x are the maximum leverage settings of the venue, not the leverage of the position. This is the single most widespread error in crypto journalism, and I have watched it mislead readers for a decade. Max leverage is a dropdown menu. Effective leverage is notional divided by equity, and it can only be computed if you know the size of the position and the size of the account. If the account holds $51.6 million of equity against a $206 million notional book, the effective leverage is 4x even though the interface says 40x. That account is not a daredevil. It is a moderately conservative institution wearing a costume.
The second possibility is that the position is genuinely 40x and the 3.5% to 7.5% figure is simply wrong — an estimate produced by someone who did not do the division.
Either way, the reader learns nothing actionable, because the reader cannot distinguish between a reckless $5 million account and a careful $50 million account. The headline is not falsifiable from the information given, and unverifiability is not an accident. It is the product. A number you can verify cannot be used to sell you something. A number you can only feel can be used to sell you anything.
The Clock That Runs While You Sleep
Now the second arithmetic, the one nobody puts in the screenshot.
Funding is charged on notional. If the reconstructed book carries something in the range of $130 million to $207 million of notional — the range implied by the two possible readings — then the daily rent is not a rounding error.
At a benign funding rate of 0.01% per eight hours, that is 0.03% per day, which on $200 million is $60,000 a day. Over a month with no price movement whatsoever, the position bleeds roughly $1.8 million. That is 35% of the reported profit, surrendered to the mechanics of holding an opinion in a crowded market.
At a crowded funding rate of 0.1% per eight hours — which happens routinely at the top of a bull leg, when every participant is leaning the same way — the daily cost is 0.3%, which is $600,000 a day on the same notional. Four days of that and the 112% return is 20%. Ten days and the whale is defending a position that has become an expense.
This is why I am allergic to the phrase "unrealized profit." It is not profit. It is a mark. It is a number that exists at a timestamp and has an ongoing maintenance bill attached to it, and the bill is denominated in the same volatility that created the mark in the first place.
At the Global Blockchain Ethics Summit in 2024, delivering a keynote on institutional entry, I argued that mainstream adoption must not dilute decentralization principles. What I did not say on stage, and should have, is that the greater risk runs the other way. Mainstream adoption does not dilute our principles. It exports our worst instrument design into pension-adjacent products and calls it access. Thirty years ago, in the early days of software engineering, we learned that a system's failure modes are inherited by everyone who builds on top of it. We are about to run that experiment again, on money.
The Bounce Was Made of Other People's Margin Calls
There is a detail in the original report that I have been circling, and it deserves its own examination: the account was, at some earlier point, close to liquidation.
Read that again, slowly, because it reframes everything. The $5.784 million is not a gain. It is a recovery. The position was underwater. It was underwater badly enough that a public report described it as near liquidation — which, for a leveraged long, means the account's equity had compressed toward the maintenance requirement, the point at which the engine begins to sell your collateral whether you consent or not.
So the headline is "trader profits $5.78 million," and the reality is "trader survives a drawdown that killed an unknown number of identical traders, then recovers."
The unknown number is the story. I have audited enough liquidation cascades to know what they look like from the inside: they are not smooth. When price approaches a dense cluster of long liquidation prices, the exchange's liquidation engine begins market-selling collateral into a book that is already thin, because everyone who would have bid is themselves being liquidated. Price falls further. More positions cross their threshold. The engine sells again. This is a reflexive loop, and it has a name in every market that has ever hosted leverage: a cascade.
Here is the part that makes me uneasy. When a cascade exhausts itself, price snaps back, because the selling was not information — it was mechanics. The violent wick you see on the chart, the one that takes out every stop and then returns to where it started, is the footprint of forced sellers being cleared. And who buys those wicks? The accounts that did not get liquidated. The accounts with the deepest collateral, the widest buffers, the largest equity relative to notional.
The whale's recovery was purchased with someone else's liquidation. Not metaphorically. Mechanically. The bounce that turned a near-liquidated account into a 112% return was, in part, the price impact of hundreds of smaller accounts being closed at the worst possible moment, creating the very dislocation from which the surviving account profited.
I am not accusing anyone of wrongdoing. Nobody chose this. It is simply what the instrument does. But I want the reader to hold the full picture, because the screenshot holds only half of it: the visible half is a winner. The invisible half is a distribution. Every leverage cascade is a wealth transfer from the impatient to the patient, dressed up as a market event, and reported afterward as though it were weather.
The Contrarian Turn: The Whale Is Not the Winner
Everything above is the conventional audit. Now let me say the thing I actually believe, which runs against both the bulls who share the screenshot and the bears who mock it.
The bulls are wrong because they think the whale won. The bears are wrong because they think the whale got lucky. Both are looking at the wrong actor.
The whale is not the winner. The whale is the inventory. On a perpetual venue, the large directional position is the thing that everyone else's business model is built around. Market makers quote around it. Arbitrageurs hedge against it. The vault earns a spread for absorbing the moments when it cannot be hedged. Funding flows toward it or away from it depending on which way the crowd leans. The whale is the reason the venue has a book. The whale is, in the most literal economic sense, a public utility with a personality.
And here is the quieter, stranger implication: the whale's floating profit is also, functionally, the venue's marketing budget. Consider what actually happened when that screenshot circulated. A platform's name was implied but not stated. Millions of impressions were generated. Inflows followed. None of it cost the venue a dollar in customer acquisition, because the customer acquired themselves. If you want to understand why high-leverage trading survives every regulatory assault and every liquidation cascade, do not look at the traders. Look at the acquisition economics. A leveraged whale in profit is the cheapest advertising unit the industry has ever invented, and it renews itself every time the market goes up.
I would be a hypocrite if I did not note the parallel to something I have written about for years. Liquidity mining programs hand out tokens to rent TVL, and when the incentives stop, the TVL leaves, because it was never there for the protocol — it was there for the yield. The same reflexivity applies to leveraged trading narratives. The volume that floods into a venue after a viral screenshot is not there for the venue. It is there for the story. And stories decay.
This is the pragmatic test I apply to any crypto narrative, and it is the one most people skip. Strip the incentive. Strip the attention. Strip the screenshot. What remains?
For a whale's leveraged account, what remains after you strip the attention is a margin balance with a maintenance requirement and a funding bill. It is not a strategy. It is a position, and positions end.
The Copy Trade Is a Decay Function
I want to address the second-order effect, because it is the one that will actually hurt people.
Copy trading is the mechanism by which a private result becomes a public liability. Its appeal is obvious: someone who appears to have a talent you lack offers to exercise that talent on your capital for a fee. Its flaw is subtler, and it is a flaw I recognized early, during my years auditing governance systems — the moment a strategy is observable, it is no longer the same strategy.
Consider what the whale's position does when ten thousand people replicate it. The aggregate notional grows. Funding turns more positive, because longs now outnumber shorts and the mechanism must punish the crowded side. Liquidation clusters thicken at predictable price levels, because everyone who copied entered at roughly the same place. The book becomes a single, densely packed herd standing on the same patch of ice.
And then something entirely predictable happens, which nobody predicts: the price dips into the cluster, the engine sells, the cluster cascades, and the entire herd is liquidated in one motion. The whale, with its nine-figure collateral buffer and its ability to add margin, survives. The copiers do not. The whale's resilience was never in the entry price. It was in the balance sheet, and the balance sheet is precisely what cannot be copied.
There is a phrase people use for this: reflexivity. Soros wrote about it decades ago in the context of equity markets, and it maps onto leveraged crypto with unsettling precision. The observation changes the observed. The copy changes the trade. The trade, once widely known, becomes a different trade with the opposite risk profile.
I spent three months in 2021 analyzing on-chain data for a thousand generative artworks in a collection called Chromie Squiggle, researching whether artists could retain moral rights over digital work through on-chain mechanisms. I wrote a manifesto then called "Algorithmic Authenticity," arguing that the chain should preserve the creator's intent, not merely the transaction history. Two hundred digital artists wrote back. The argument I made then applies here, though the creators are traders and the art is a position: a chain preserves record. It does not preserve meaning. Meaning has to be carried by humans, and humans are exactly what get copied over.
What a copy trader is actually copying is not a decision. It is a decision's residue, visible only after it has already been priced in.
The Thirty Seconds of Boredom That Kills Accounts
Let me get concrete about the failure mode, because abstraction is how people avoid learning.
At 40x, a long position's liquidation price sits approximately two to two and a half percent below entry, once you account for the maintenance margin requirement. At 25x, it sits approximately three and a half to four and a half percent below. These are not scenarios. They are the ordinary daily range of both assets. A single macroeconomic print, a single whale wallet moving, a single exchange outage that delays your ability to add margin by ninety seconds — any of these is sufficient.
The source report suggested that a correction of roughly 3.5% to 7.5% would trigger partial liquidations. I computed a tighter band and I will explain why the difference matters. If the true distance to liquidation is 2%, then the window between "safe" and "gone" is smaller than the latency of a mobile order confirmation screen. If it is 7.5%, the account has substantial idle collateral and is nowhere near as aggressive as advertised. There is no version of this where the reader can act on the number, because the reader does not know which version they are in.
The absence of the entry price is the whole problem, and its absence is not a page-layout accident. Entry price, liquidation price, venue, and collateral source are the four facts that determine whether a leveraged position is a considered bet or a countdown timer. The report contains none of them. It contains a profit figure and a percentage, which are precisely the two numbers that can be published without revealing risk.
If you have taken nothing else from this essay, take this: any report of leveraged profit that does not disclose entry, liquidation price, and venue is not information. It is advertising with a charts aesthetic.
The ethical dimension of this is not abstract. In my 2017 work auditing a DAO that had inherited TheDAO's burden of proof, I spent twelve weeks reviewing 150,000 lines of Solidity and identified 42 logic flaws that were not syntax errors at all — they were failures of trust assumption, places where the code believed something about human behavior that was no longer true. Code is law only when it aligns with human values. A liquidation engine is perfectly lawful. It executes exactly what it promises, in milliseconds, without malice or discretion. And it can still destroy a family's savings because a number on a screen crossed a line that nobody explained to them.
The technology is not the problem. The explanation is.
What Would Actually Be Useful
I have spent this entire piece dismantling a headline. Let me spend a little of it building something.
If I were writing the report that should have been published, it would contain four elements.
The first is effective leverage, computed as notional divided by account equity, disclosed alongside the account's maintenance margin requirement. This one number transforms a leveraged position from a mood into a measurable object.
The second is distance to liquidation, expressed both in percentage terms and in dollars, with a note on how much margin would need to be added, and at what price, to hold the position through a two-sigma move. This is what a risk desk produces. It is not difficult. It is just unflattering.
The third is the cost of carry: current funding rate, trailing seven-day average funding, and the cumulative funding paid since the position opened. A position that is up 112% gross and 40% net is a very different animal from one that is up 112% net, and readers deserve to know which they are admiring.
The fourth is the venue, named, with a one-sentence statement of its oracle design and its governance override capability. If a validator set can vote to change a settlement price, that belongs in the disclosure. It is not a scandal. It is a property, and properties belong in the specification.
None of this is exotic. All of it is standard practice in every regulated derivatives market on Earth. The fact that the crypto industry treats it as an imposition rather than a baseline tells you how young we still are, and how much of our apparent sophistication is borrowed costume.
The Part Where I Admit I Could Be Wrong
Halfway through writing this, I went back and checked my own math, because the thing I trust least in the world is a clean argument.
If the 112% figure refers to return on total account equity rather than posted margin, and if the account is very large, then the reconstructed notional shrinks, the funding burden becomes a footnote, and the whale is simply a well-capitalized institution running a modest directional book under a flashy leverage setting. In that reading, the headline is misleading in the way a carnival barker is misleading: technically true, aesthetically inflated.
If the 112% refers to posted margin at genuine 40x, then the account is one three-percent candle away from the same fate as everyone else in the cluster, and I would not wish that on anybody, least of all on the copy traders who followed them in.
I cannot tell which world I am in, and that inability is my point. I audit code for a living. I have spent twenty-six years learning to distinguish a system that is safe from a system that merely looks safe, and the longest lesson of that career is this: the systems that fail most spectacularly are the ones that never made anyone uncomfortable, because nobody bothered to look.
I am not asking anyone to be afraid. Fear is a poor analytical tool; it produces the same errors as greed, just with worse posture. I am asking for the four numbers. Entry. Liquidation. Funding. Venue. Four numbers, and a screenshot stops being a story about a stranger and becomes a document you can read.
The Truth Layer Does Not Flatter Anyone
I finished a six-month open-source project last year, building a verifiable training dataset on-chain to establish data provenance and reduce algorithmic bias. I went in with genuine doubt about whether any of it mattered — whether the industry building the settlement rails for a machine-intelligence economy had any real claim on truth, or was just the loudest room in the building. Three researchers and I published a final report arguing that a chain can serve as a truth layer for AI. I stand by it. I also stand by the corollary I did not put in the abstract: a truth layer only helps people who are willing to be told something they did not want to hear.
That is the entire problem with the leverage narrative, and it is not a technical problem. It is a human one. The four numbers exist. Anyone with a wallet address and an API key can find them. The information is available, and the appetite is not. What people want from a $5.78 million screenshot is permission, not data. And permission is the one thing a settlement layer has never been able to issue.
There is a chart I keep coming back to, from the worst week of the 2022 collapse. Not the price chart — the open interest chart, which fell so far and so fast that the market's leverage ratio reset to levels not seen since 2019. Everyone described it as destruction. It was closer to detoxification. The industry got healthier because it got smaller, and it got smaller because the leveraged were forcibly removed. Six months of that, and I wrote a 30,000-word analysis in self-imposed isolation, rejecting consulting offers because I wanted to say what I believed without it costing me a retainer. That period produced the most honest writing of my life, and it took a bear market to make it possible.
We are in a bull market now. Honesty is more expensive. So it costs less to produce, and there is less of it. That is not a coincidence.
What I Would Watch Instead
If you find yourself drawn to a screenshot like this one, I will offer an alternative practice, and it will strike you as less exciting, because it is.
Watch funding. Not the price. The funding rate on the venue in question, and its trajectory over seventy-two hours. A rate that climbs and holds above 0.05% per eight-hour interval is a market telling you the crowd is leaning long and paying for the privilege. Crowded funding is the most reliable leading indicator of a cascade that exists, and it has the virtue of being public, free, and boring.
Watch open interest against price. Rising price with rising open interest is new money arriving. Rising price with falling open interest is positions being closed into strength — a rally running on fumes. Falling price with rising open interest is the most dangerous configuration in the market: new shorts entering against a still-motivated long base, the setup where cascades get violent in both directions.
Watch the liquidation clusters, not the line. Price is drawn to liquidity the way water is drawn downhill, and the largest pool of resting orders on a leveraged venue is the collection of liquidation prices. When a dense cluster sits two percent below spot, that is not a support level. It is a magnet.
And watch your own position size with a ruthlessness you would apply to someone else's code. The only risk metric that has ever mattered is the distance between your liquidation price and the ordinary daily range of the asset you are holding. If that distance is smaller than the range, you are not trading. You are waiting.
A Closing Thought, Offered Without Summary
I do not know whether the account in the screenshot belongs to a person I would admire or a person I would warn my students about. I suspect it belongs to someone who has done this long enough that the number no longer feels like anything — which is its own kind of loss, and one that nobody publishes.
What I know is this. The industry I have spent my adult life defending, the one whose values I still believe in and whose excesses I have spent six years documenting, has produced a financial system of extraordinary elegance and extraordinary cruelty, and it has placed them so close together that most participants cannot tell which one they are standing on. The elegance is real. The composability, the permissionlessness, the ability of a twenty-two-year-old in Lagos or Lisbon to access instruments that were once reserved for trading floors — all of that is real and all of it still moves me.
But elegance is not safety, and permissionlessness is not kindness, and a screenshot of a stranger's floating profit is not an invitation. It is a mirror, and what it shows you is what you already wanted to believe.
The lever does not care what you believe. It only measures the distance to your liquidation price, in basis points, in real time, while you are reading something else.
And the truth layer — the one we spent a decade building, the one that will not lie, the one that will still be there when the funding rate resets and the screenshot scrolls off the feed — the truth layer does not flatter anyone. That is not its defect. That is the only reason it is worth having.
I remember a line from the 42 flaws I found in that DAO's code, all the way back in 2017. The flaw was never in the arithmetic. The arithmetic was always correct. The flaw was in what we assumed about each other, and how long we assumed it would last.
It lasted until the price moved.
