Not because you set it badly. Because the level it was protecting has moved. Funding accrues, you add margin, another position gains or loses — and your liquidation price shifts underneath it. The stop you placed once does not follow.
And that is only the first problem. Your exchange shows one liquidation price per position; in a real crash that number is useless, because everything falls at once and the price does not glide down — it jumps over the levels where other people are being liquidated. Check both on your own account, below, in about ten seconds.
| Account | Asset | Side | Position | Leverage | Room |
|---|
| Asset | At risk | We see | Share of that |
|---|
Read-only. We never ask for a key, a signature, or a connection. No idea where to start? Show me the most exposed account on the exchange right now.
One message when your account gets within 5% of its liquidation level. Nothing the rest of the time — no daily summary, no newsletter. We never ask for a key or a signature.
Each column is a price scale for one of your positions. Your liquidation level is the red line. The blue bands are other traders' liquidation levels, on the side that threatens you — below the price if you are long, above it if you are short. Those are the points where a block of forced orders hits the market and the price gaps instead of sliding. The gold marks are where a protective order would sit: in front of the waves, not inside them.
Every bar is a real day: how far the whole market moved, measured as the median across 177 assets over the last two years, in the direction that actually hurts this account. The gold line is the general move that liquidates your account.
| Asset | Side | Notional | Price | Liquidation | Room | Exit cost | Forced exit |
|---|
Exit cost is what it costs you in slippage to close the position calmly. Forced exit is what the same position costs when the exchange closes it for you, into a falling market. The gap between those two columns is the entire reason this product exists.
Three out of four positions I close were minutes or hours from liquidation. Left alone, the exchange takes the entire margin — not part of it, all of it. Every one of these is counted against a liquidation Hyperliquid itself confirms, and the table grows with the archive. Nothing here is typed into the page.
| Room to liquidation | Snapshots | Liquidated | Risk | vs baseline |
|---|
Counted on 3,502 liquidations confirmed by the exchange itself — each one read back from the fills Hyperliquid marks as a liquidation, on the exact position and to the second. Not inferred from positions we saw disappear. Fifteen times the background rate. And a position sits inside that last 1% for a median of 17.8 hours before it dies — which is the entire reason acting in time is possible at all.
An alert cannot help you at four in the morning, and it cannot help you when the price gaps over your level in seconds.
Places a ladder of orders at the exchange, sized so that each one moves your liquidation out of reach. They rest there and cost nothing until a price is reached. If my server dies, they still work.
Move your money. The authorisation you sign is an agent key: on Hyperliquid it can place orders and can never withdraw. That is a property of the protocol, not a promise of mine. You revoke it in one click.
At the point where I act, roughly one intervention in four turns out unnecessary — the price comes back and you would have survived. Those positions recover a median of 2.7% of room, so being closed early costs you very little. Being liquidated costs you the entire margin.
And there is a ceiling. It is not a percentage of your capital — it is your own profit. I only spend what the account has made. Once my orders have cost you what you were up, I stop placing new ones. While you are below where you started, I take nothing at all. Short of the cliff below, my protection cannot leave you worse off than you arrived — and that is a sentence a stop-loss can never say, because a stop-loss does not know when to stop.
The line I measure from is what the account was worth at the start of the month, and it resets each month at whatever the account is worth that day. So a flat month is a month where I only step in at the cliff.
And a second ceiling, on each order rather than on the month: no single order I place can cost you more than 2% of your stake in slippage. If a rung would cost more, I do not drop it — I make it smaller. That is the difference between a protection that is dosed and one that is switched off.
This is the fix for the worst thing this kind of system can do to you. The old ceiling was a share of the position, not of your account — and on a book carrying thirty-five times its capital, 2% of the position is 70% of everything you have. Measured across three weeks of order books: accounts killed by the protection went from five to zero, and not one winner was lost.
The ceiling has one exception, and it is the important one: it never applies when you are about to be liquidated. Fees run high in a crash precisely because that is when the work is being done. Stopping there would mean abandoning you at the exact moment you hired me for.
| Money already sitting within 2% of its liquidation | Hours observed | Liquidations in the next 6 hours |
|---|---|---|
| Under $0.5M | 442 | 3.2 |
| $0.5M – $2M | 193 | 4.2 |
| $2M – $10M | 164 | 9.7 |
| $10M – $50M | 58 | 10.4 |
Accounts do not die alone. A liquidation sells into the book, the price moves, and it takes the next one with it — liquidations arrive in clusters 4.2 times tighter than chance. On 22 August at 05:00 UTC: 22 on XRP, 21 on SOL, 17 on BTC, 15 on ETH, inside the same hour. So my threshold is not fixed. When the powder keg on your asset is loaded I take you out earlier, because the wave is coming. When it is empty I leave you alone. Computing this requires re-reading leveraged accounts across the exchange, every few minutes, for months — 40 million position snapshots so far — which is why your stop-loss cannot do it, and neither can anyone watching only your account.
A position can only be closed if it can be closed at a sane price. On an illiquid pair it cannot, so no rung is placed there — and you should know that before you sign, not after.
Measured on 1,596 real accounts, using the thinnest order books of the three weeks tested, so this is the bad case: two accounts in three can be covered at 99% or more, and 86% can be covered at half or more. The median account is covered in full, whether it holds one pair or forty.
The exception is worth stating plainly: 80 accounts out of 1,596 — one in twenty — cannot be covered at all. Almost all of them hold a single illiquid position. There is no rung I can place for them at a price worth paying, and I would rather say so here than take their fee.