A wallet deposits $45. It is a real wallet — one of roughly 400,000 that trade Hyperliquid perpetuals through six consumer apps: Phantom, MetaMask, Rabby, Trust Wallet, Invo and fomo. Each of those apps connects its users to the exchange through a builder code and earns a fee on every trade. The traders, as the on-chain analytics firm Hyperdash documented this month, are down $280 million all-time. About $126 million of that was lost just since November 10th, 2025, as the apps scaled. Typical deposits at five of the six apps: $30 to $60.
The figure that should stop everyone is not the total. It is the composition. Slightly more than half of the $126 million was not lost to the market at all. It was paid in fees.
Each app adds its own builder fee on top of the exchange's base fee, so a $100 order costs between 6.5 and 14.5 cents depending on the app. That sounds like nothing, until you look at the turnover it invites. Invo's users deposited $12.4 million and traded $5.8 billion — the average deposit changed hands hundreds of times, and on every pass the fee schedule took its slice.
This is the part the trading industry has always preferred to keep quiet: for a small account that trades often, the broker's cost is a position in its own right. On BTC — the market these users trade most, close to half of all app volume — fees alone exceeded what users made on the trades they closed. They did not lose because they were wrong. They lost because they were active, and activity has a meter running. The exchange was not even the main counterparty in their failure; their own app's revenue model was.
The business side worked brilliantly. Monthly active traders across the six apps more than tripled from November to September, led by Invo and fomo, which each grew from almost nothing to tens of thousands of traders within months. The apps are a genuine acquisition channel — more traders, more volume, more fees every month.
Retention is where the design shows its teeth. About half of each month's active traders are trading through the apps for the first time, and only around one in seven new traders is still trading three months later. Growth did not come from traders who stayed. It came from continuously replacing the ones who left. An industry that loses most of its customers every quarter and keeps growing is not overcoming churn. It has made churn the product.
The dataset is predominantly small retail accounts, and the distribution tells its own story. At five of the six apps, the typical wallet deposited between $30 and $60. Rabby is the exception — fewer and considerably larger accounts — and its users are the most likely to be profitable. That single contrast is worth sitting with: the variable that moved the profit needle was not market selection, not timing, not which app's interface was smoother. It was account size, because a larger account turns over less relative to its own balance and pays less of itself away in fees. The smaller the account, the faster the meter eats it.
Activity concentrated where retail already feels at home: BTC took close to half of all app volume, followed by the newer HIP-3 markets — oil, silver, gold — which barely traded through these apps before January and are now the second-largest category. The migration into commodities perps happened in weeks, and it cost users more per dollar traded than BTC and ETH taken together. New markets did not broaden the audience's competence; they gave the fee schedule fresh surface area.
Here is the structural fact underneath all the numbers: the app is paid per fill, not per outcome. Every dollar the trader pays in fees is a dollar the app earns. The trader's cost line and the platform's revenue line are the same line, read from opposite ends.
Nothing in that arrangement requires malice, and the author of the study — an analytics firm embedded in the Hyperliquid ecosystem — is careful to say the apps deliver exactly what builder codes were designed to deliver. That is precisely what makes it worth understanding. A revenue model does not need bad actors to eat an account. It needs only that its income grows with the behaviour that most reliably destroys the account: overtrading, high fee drag, liquidations. The incentive does not point at retention; the study's own numbers show it — a business that triples its active traders while keeping one in seven has no financial reason to help anyone get better.
The uncomfortable comparison is that this is not exotic. Independent research on the same exchange found only about 29 percent of directional traders on its native frontend profitable — clustering near the 75-85 percent loser rates that leveraged CFD brokers have been legally required to disclose for years. Hyperliquid simply makes the same statistic visible on-chain, wallet by wallet, in public.
One in seven sounds like an indictment of the traders. It is better read as an indictment of the setup. A first-time trader with a $45 deposit, trading BTC through an app that earns on every fill, is not participating in a market — they are feeding a cost machine with no edge and no system to carry. The three-month exodus is what happens to capital that arrives without rules in a venue paid to accelerate its motion.
Which inverts into the whole discipline argument from the other side of the glass. Everything that would have kept those traders in the game is the opposite of what the fee schedule rewards: fewer trades, pre-committed risk, a rule set carried by automation so the activity level is decided before the session and not negotiated with the app mid-trade. The edge those wallets needed was never information. It was slowness.
Same arithmetic governs us, so plainly. Our automation runs on fixed rule sets executed by the machine — the trade frequency is a parameter, not a mood. Stops and limits are placed before any position exists. A monitor watches the engine's behaviour for drift from its verified baseline and cuts its authority when drift appears. Parameters are localised per market and broker, because a cost structure that is survivable on one instrument is fatal on another. In build, not yet shipped: an analysis layer that reads a trader's own closed-trade history and reports what it shows — including the one line the Hyperdash study proved matters more than any market call: how much of the loss was the market, and how much was the meter.
None of it predicts. All of it contains. The apps in this study grew by replacing their traders; a system survives by making its trader worth keeping.
The market changes. The rules don't.
