Every Hyperliquid liquidation, fill by fill, from the node fill stream: coin, side, price, size, USD notional and whether it cleared in the market or hit the backstop. These are exact fills, not estimates. Below are the newest; the full tape and incremental bars are on the Pro API.
| Time | Coin | Side | Price | Size | USD | Method |
|---|---|---|---|---|---|---|
| 11:47:18 | NEAR | short | 4.661 | 2.4 | $11 | market |
| 11:47:18 | PUMP | long | 0.00421 | 2298 | $10 | market |
| 11:44:17 | ETH | long | 2,718.5 | 1.3168 | $3.6K | market |
| 11:44:17 | ETH | long | 2,718.5 | 0.0287 | $78 | market |
| 11:44:14 | ETH | long | 2,718.7 | 0.008 | $22 | market |
| 11:44:14 | ETH | long | 2,718.7 | 0.0043 | $12 | market |
| 11:44:14 | ETH | long | 2,718.7 | 0.3148 | $856 | market |
| 11:44:14 | ETH | long | 2,718.7 | 0.0195 | $53 | market |
Grain: each tape row is one exact liquidation fill (grain.kind: event). Many fills share a millisecond, so page with cursor, never start = last.time + 1. For flow you can add up, /hl-liquidation-bars buckets the same fills into closed long/short USD bars (grain.kind: bar, sum_ok: true).
The 1h/4h/12h/24h totals on /liquidations and /market-intelligence/liquidations are different: rolling-window levels (grain.kind: rolling_window, sum_ok: false). Don't sum or difference successive polls; use the bars for flow.
next_cursor
Every liquidation fill: time, coin, side, px, sz, usd, method, mark_px, filterable by coin and time range. Carries grain and coverage blocks so agents know what a row is and how far back it goes.
The same fills bucketed into long/short USD bars per interval: incremental flow you can sum, join to klines and backtest. This is the correct series for cascade studies; the rolling 24h board is not.