Liquidation Hunting on Hyperliquid: How It Works and How to Bot It

Hyperliquid's fully on-chain order book makes liquidation hunting viable in a way that CEX perp markets simply do not allow. Every open position, every leverage ratio, and every oracle update is visible in real time. If you know where to look and how to act faster than the liquidation engine, you can enter a trade moments before a cascade prints a wick and exit into the recovering spread — repeatedly, with controlled risk.

How Hyperliquid Liquidations Actually Trigger

Hyperliquid uses a mark-price oracle derived from a weighted median of external index prices. When mark price crosses a position's liquidation threshold, the protocol sends that position to the liquidation engine, which first attempts to close it via the order book at market. If the order book cannot absorb it, the position is taken over by the Insurance Fund and closed against a socialized-loss pool.

The key mechanic: the engine does not try to minimize slippage. It hits bids — all of them — until the position is closed. On a thin book with a cluster of longs stacked at a round-number support, a single engine call can clear three or four price levels in under 100 milliseconds. That cascade is the wick you are hunting.

Liquidations are deterministic once you know the price threshold. The variable is timing: oracle updates arrive roughly every 3 seconds, so a position that is 0.4% above its liquidation price right now will survive until the next oracle tick crosses that gap.

Reading the Leverage Map

The Hyperliquid L1 exposes all open interest data via its info endpoint (POST /info with {"type": "clearinghouseState", "user": "..."} for individual accounts, or the aggregate openInterest and meta endpoints for market-wide views). Third-party dashboards aggregate this, but if you are building a bot you want raw data, not a scrape.

What you are looking for:

• Notional clustering: large open-interest concentrations within 1-3% of current mark price, especially on 10x-25x leverage where maintenance margin is 0.5-1%
• Imbalance direction: if longs outnumber shorts 3:1 at a given price band and price is falling toward that band, the liquidation engine will be a net seller
• Thin book beneath: use the L2 snapshot ({"type": "l2Book", "coin": "ETH"}) to measure bid depth between current price and the cluster. Shallow depth means the engine will move price further before filling

The setup you want: a large long cluster sitting 1.5-2.5% below spot, with less than $200k in bids between them and current price on a mid-cap asset like SOL or ARB.

Structuring the Entry

Do not try to predict when the oracle crosses. Instead, set a trigger band: if price enters within 0.6% of a liquidation cluster, open a short. The thesis is that either (a) the cluster liquidates and you ride the wick down, or (b) price bounces before the cluster and you scratch out near zero. The payout is asymmetric.

Position sizing matters more than entry precision here. Because the wick duration is typically 200-800ms, you are not scaling in. You enter a fixed notional — usually 0.3-0.8% of capital — and exit entirely on the other side of the wick. On Hyperliquid you can place a limit order to exit at mark price + N% simultaneously with your entry; the latency between mark-price update and order-book execution gives you a real fill.

Typical captured spread on a clean cascade: 0.4-1.2% on the position notional, before fees. Hyperliquid charges 0.035% taker on the way in and out, so net on a 0.6% gross capture is roughly 0.53%. Do that 15 times a week on $10k notional and the math is material.

Risk Controls You Cannot Skip

False signals are the main killer. The bot must handle:

• No-cascade liquidations: the engine fills quietly against resting liquidity and price barely moves. Your short is offside. Hard stop at 0.25% adverse move. No averaging down.
• Gap opens after funding: funding resets at 08:00 UTC and can cause a sharp mark-price jump that triggers your trigger band without a corresponding spot move. Filter: only enter if the spot index (not just Hyperliquid mark) is within 0.3% of the same level.
• Insurance Fund intervention: if the IF absorbs the position rather than hitting the book, price impact is suppressed. This is rare but happens on very large positions. The tell is that open interest drops but price does not move — close immediately.
• Latency slippage on exit: your exit limit needs a fallback market order if unfilled after 1.5 seconds. Without this, you hold a directional short while the wick recovers.

Log every trade with the trigger OI cluster, the entry/exit prices, the actual wick depth, and whether the cascade was engine-driven or organic. That dataset is how you tune the trigger band over time — what works on SOL at 10am UTC is different from ETH at 3am UTC.

Execution Infrastructure

The bot architecture is straightforward: a polling loop on the info endpoint at 500ms intervals, a position-map builder that updates on each tick, and an order-placement function that hits the exchange API with pre-signed L1 transactions. Hyperliquid's API uses EIP-712 typed data signatures, so your key management needs to be local and fast — no remote signers in the hot path.

If you are running this in Python, httpx with asyncio will handle the polling. The order-placement leg needs to complete within 300ms of trigger detection; if your signing overhead exceeds that, move to Rust or Go for the execution module while keeping Python for the analytics layer. The bots we build at TierZero follow exactly this split: a typed-message signer in Rust called via subprocess, orchestrated by a Python strategy layer that handles position tracking and risk state.

What This Strategy Is Not

It is not arbitrage. You are taking directional risk, however short-lived, and the edge comes from information (the leverage map) rather than a locked spread. That means drawdown periods exist — typically when markets are trending cleanly upward with low leverage accumulation, there are no clusters to hunt and the strategy sits flat for days. That is fine. Forcing trades when the setup is absent is how liquidation-hunting bots blow up.

Run it as one component of a multi-strategy system, not the whole book.

---

If you want a production liquidation-hunting bot on Hyperliquid with proper risk controls and execution infrastructure already built, talk to us.