One row per fill leg on Hyperliquid perpetual futures, covering both sides of every match across first-party and HIP-3 builder-deployed markets.
hyperliquid.perp_trades is the fill table for Hyperliquid perpetual futures, venue-wide: first-party markets (coin = BTC) and HIP-3 builder-deployed markets (coin = xyz:TSLA) alike. Grain: one row per fill leg — both sides of every match — keyed on (block_date, coin, trader, oid, tid), with tid shared by the two legs of a match. Spot and HIP-4 outcome markets are out of scope.Fills are decoded from the full node fills stream rather than from taker order actions (fills_raw), which misses every engine-generated fill — TWAP child slices and liquidation executions — worth 5-30% of volume. History starts 2025-07-27 and covers every coin that has traded, at millions of fill legs per day.
Table schema
Column
Type
Description
block_month
DATE
Month of block_date, for month-grain grouping. Not the partition key
block_date
DATE
Partition key, and the only column that prunes
block_time
TIMESTAMP(3) WITH TIME ZONE
Block time of the fill
block_number
BIGINT
Block height
event_index
INTEGER
Position of the leg within the block; the two legs of a match sit at consecutive indices
perp_dex
VARCHAR
hyperliquid for first-party markets, otherwise the builder DEX code (xyz, flx, para, …)
coin with the dex: prefix stripped. Not unique across dexes — never join on it
asset_id
BIGINT
Numeric asset id used by the raw Hyperliquid action tables. NULL when a market traded before the registry poller recorded it
trader
VARBINARY
The account on this leg, as raw bytes rather than a hex string
is_taker
BOOLEAN
Aggressor flag. Exactly one leg per match has it set
side
VARCHAR
buy or sell from this leg’s perspective
dir
VARCHAR
Raw Hyperliquid direction string, e.g. Open Long, Close Short — the position effect, not the order side
fill_type
VARCHAR
Normalised leg role: order, twap, liquidation, adl, settlement, vault_netting
start_position
DOUBLE
Signed position size immediately before this fill
price
DOUBLE
Fill price, in the market’s quote token (see fee_token)
size
DOUBLE
Fill size in base units
notional_usd
DOUBLE
price * size. Sums to ~2x reported volume unless you filter is_taker
closed_pnl_usd
DOUBLE
Realised PnL on this leg; non-zero on closes
fee_usd
DOUBLE
This leg’s fee, in fee_token. Negative on maker rebates
builder_fee_usd
DOUBLE
Front-end / builder-code fee share, populated on ~6% of legs
deployer_fee_usd
DOUBLE
HIP-3 deployer fee share
fee_token
VARCHAR
Quote / collateral token the prices and fees are denominated in. Not always USDC
priority_gas
DOUBLE
HYPE spent in the priority-gas auction. Sparse
tid
BIGINT
Match id, shared by both legs — the pairing key
oid
BIGINT
Order id, globally unique
cloid
VARCHAR
Client order id where the trader set one
builder
VARBINARY
Builder (front-end) address, populated on ~6% of legs
twap_id
BIGINT
TWAP order id, set only on the taker leg. Group by it to reconstruct an executed schedule
liquidation_method
VARCHAR
market or backstop. Set on both legs of the match — use fill_type for the role
liquidated_trader
VARBINARY
The account that was liquidated. Also present on both legs
liquidation_mark_price
DOUBLE
Mark price at liquidation
_updated_at
TIMESTAMP(3) WITH TIME ZONE
Build timestamp. Legs are immutable, so in practice when the leg was first written
Bound block_date on every query. It is the partition key and the only column that prunes: block_month is a plain column, so filtering it alone reads every day in the month, and nothing prunes within a day — a coin filter still scans the full partition. An unbounded scan reads the whole multi-billion-row history.
Both legs are here, so filter is_taker for volume
SUM(notional_usd) over this table is ~2x exchange-reported volume, because every match contributes a taker leg and a maker leg. Filtering is_taker counts each match once. For pre-aggregated volume, use perp_market_metrics_hourly or perp_market_metrics_daily, which are the volume source of truth.
-- Traded volume per market, single-countedSELECT block_date, coin, SUM(notional_usd) AS volume_usd, COUNT(*) AS tradesFROM hyperliquid.perp_tradesWHERE block_date >= current_date - INTERVAL '7' day AND block_date < current_date AND is_taker AND fill_type NOT IN ('settlement', 'vault_netting')GROUP BY 1, 2ORDER BY 3 DESC
Do not filter is_taker when the question is per-account: at account grain each leg is one account’s own participation, which is also why perp_accounts_daily aggregates both legs. The same applies to closed_pnl_usd — maker-side closes carry realised PnL too, and a taker-only filter drops roughly 40% of it.
fill_type
fill_type is the leg’s role, normalised from dir, liquidation_method, is_taker and twap_id.
Value
Leg role
order
A voluntary order fill, including the resting order hit by an ordinary liquidation
twap
The leg belongs to a TWAP child slice; always the taker
liquidation
This leg’s account was force-closed for undercollateralisation
adl
This leg’s account was the profitable side force-closed by auto-deleveraging
settlement
Engine force-close of all open positions when a builder DEX delists a market: one price, one timestamp, zero fees
vault_netting
Vault position netting (dir = 'Net Child Vaults'), not a trade at all
settlement and vault_netting are administrative and belong in no volume or activity measure. liquidation and adl are genuine risk transfers at market prices, but they are forced, so include them only when the question covers forced flow.TWAP legs inflate fill counts far more than volume — one parent order becomes many slices. Count DISTINCT twap_id for order-level figures.
Liquidation columns mark the match, not the role
liquidation_method and liquidated_trader are stamped on both legs of a liquidation match. Counting rows where liquidation_method IS NOT NULL therefore doubles the answer: measured 2026-08-12, 17,044 legs carried a method while only 8,522 accounts were liquidated. Use fill_type = 'liquidation' for the forced leg.
The roles depend on the method. On market (ordinary) liquidations the liquidated account is the taker and its counterparty is a plain order leg. On backstop (auto-deleveraging) the roles invert: the bankrupt account is the maker leg with fill_type = 'liquidation', and the profitable account force-closed against it is the taker, with fill_type = 'adl'.
side is the order direction, dir is the position effect
The two legs of a match always carry opposite side values. dir is not symmetric that way: it describes what the fill did to this leg’s own position — Open Long, Close Short, Long > Short — so a Close Long fill has side = 'sell'. Use side for order flow, dir or start_position for position intent. Perp scope carries exactly 13 dir values, verified across full fill history; an unrecognised value lands in fill_type = 'order' until the model is taught it.
Fees
fee_usd is negative on maker rebates, so summing it across all legs yields net protocol revenue; add is_taker for gross taker fees. builder_fee_usd is populated together with builder on the ~6% of legs routed through a front-end, and deployer_fee_usd only on HIP-3 markets.All three are denominated in fee_token, which is not always USDC — a HIP-3 DEX picks its own quote token, and USDC, USDT0, USDH and USDE all appear across history. They are USD-pegged, which is what keeps the _usd column names honest, but a fee on a USDH market is paid in USDH.
Not in this table
Curated classification — market_category, asset_class, asset_type, underlying_ticker, builder_name — is deliberately kept off this table so a reclassification does not rewrite billions of rows. Join perp_market_details on coin for it, along with the market-level max_leverage, margin_tiers and margin_mode.Per-fill leverage and margin mode are absent by design: both are a per-(user, asset) account setting rather than a property of a fill, and the venue-wide models do not reconstruct them.