Perpetual futures history recorded event by event rather than sampled: the order book as the venue publishes it, the best bid and offer on every change, every trade, and the ticker stream that carries mark, index, open interest and funding.
Two venues: Hyperliquid, every perp since 2025-01-25 rebuilt from the chain's own order-event stream, and Polymarket perps, captured live. Parquet, sold per day with a coin or instrument filter. No feed to run, no snapshots to stitch, no funding history to scrape from a chart.
Polymarket lists 83 perpetuals and Hyperliquid has a counterpart for essentially all of them: the crypto names natively (BTC, ETH, SOL, HYPE, kPEPE), and the equities, indices and commodities (NVDA, TSLA, TSM, GOLD, the S&P 500, Brent) on Hyperliquid's xyz HIP-3 dex. Both feeds carry the venue's own timestamps, so the question a cross-venue dataset exists to answer — who prints first, by how much, and does it hold on unseen days — is a join, not a project.
We ran that join across the six weeks both captures overlap: Research note 006: Polymarket perps print 250–500 ms before Hyperliquid on source clocks. The note states its clocks, its holdout, and what it does not claim about executability.
Both venues on one account: a plan bought for Hyperliquid at L10 ($49/mo, 10 levels) or L25 ($99/mo, the full 25-level book): every perpetual, native and HIP-3, plus the spot pairs, with Polymarket perps as a $29/mo add-on. Or a Polymarket plan next to it (every one includes Polymarket perps, L10 and L25 in full): each venue is its own subscription and they add up.
Coverage dates are our own capture-start dates, and capture has been continuous since. Counts are read from the catalog when this page renders.
235 perpetuals (plus 346 spot pairs and 314 HIP-3 markets), 641 days on disk, 2025-01-25 → 2026-10-27, 6914 GB of parquet across three tables: top of book on every change, the 25-level book on every change with resting-order counts, and every fill with taker wallet, fee and realized PnL — reconstructed from the chain's own order-event stream, one row per block. Full 25 levels on every plan.
91 instruments across crypto, US equities, indices and commodities. 89 days on disk, 2026-07-31 → 2026-10-27, 927 GB of parquet across six tables: event-driven BBO, 20-level books every ~100 ms, the full ladder every ~15 s, every trade, tickers with funding and open interest, and the instruments dictionary.
One parquet per table per instrument per UTC day. Every event table carries the venue's own timestamps and our nanosecond receive clock, and the dictionary ships with every pull.
perps_bookThe websocket ladder, 20 levels a side, as a complete snapshot roughly every 100 ms. The venue publishes no delta stream, so every row is a whole book; about a quarter of consecutive frames are byte-identical and are kept as sent.
perps_book_deepThe full ladder from the REST book endpoint, polled about every 15 seconds — deeper than the 20-level websocket book (deepest observed ~71 levels) and carrying the venue's own sequence and timestamp. A separate table, never merged into perps_book.
perps_bboTop of book on every change: bid/ask price and size, mid and spread, straight off the venue's bbo channel. ~17 messages a second per instrument, none of them redundant.
perps_tradesEvery fill the venue printed: taker side, price, size, venue trade id and the settlement transaction hash once mined. The venue is young and thin — tens of thousands of trades a day across all instruments — so the book and ticker tables carry most of the information.
perps_tickersIndex, mark, last and mid price, open interest, the current funding rate and the next funding time, roughly every 100 ms per instrument. This is the ONLY place funding lives: the venue has no market-wide funding channel.
perps_instrumentsThe contract specifications and the iid → symbol mapping every other table is keyed on: category, base/quote asset, max leverage, tick/lot decimals, min notional, liquidation fee, risk tiers. Snapshotted every five minutes. Included with every pull.
Most perps venues publish an incremental book: a snapshot, then deltas you apply in sequence, which means any gap in the stream silently corrupts every book after it. Some publish full snapshots instead, which cost more bytes and can never drift. We record what the venue actually sends, say which it is per venue, and never reconstruct one from the other after the fact.
Funding and open interest are the perps-specific series a spot dataset does not have, and they are the ones most often reduced to an hourly chart after the fact. Here they arrive as ticks on the ticker stream, so you can see the rate move inside the funding interval and line it up with the book on the same clock.
Every bundle README states the caveats for your exact range: on Polymarket perps the first day is partial (capture began 15:44 UTC on 2026-07-31) and KPEPE-USD is absent 2026-08-04 to 2026-08-05.
A row per event from a perpetual futures exchange rather than a row per interval: the order book each time the venue publishes it, the best bid and offer on every change, every trade as it prints, and the ticker stream that carries mark price, index price, open interest and the funding rate. Nothing is sampled on a timer or bucketed into candles before it reaches you.
It depends on the venue, and we say which. Hyperliquid's books are reconstructed from every order event the chain executed, so each row is the full 25-level book after a block in which it changed — finer than any snapshot feed, with no duplicate frames. Polymarket perps publishes no incremental stream, so every book row is a complete 20-level ladder at roughly 100 ms, with a separate table holding the full ladder from the REST endpoint every ~15 seconds; about a quarter of consecutive frames are byte-identical and are kept as sent.
From the ticker stream, on every row alongside mark, index, last and mid price and the next funding time. On Polymarket perps there is no market-wide funding channel, so the ticker table is the only place funding history exists.
Two. Hyperliquid — every perpetual the venue has listed since 2025-01-25 (BTC, ETH, SOL, HYPE and a few hundred more, plus its spot pairs and HIP-3 builder dexes), with the book on every change and every fill, reconstructed from the chain's own order-event archive. And Polymarket's perpetual futures exchange, captured on two independent collectors since 2026-07-31: 67 instruments across crypto, US equities (NVDA, TSLA, MSTR, COIN and others), the S&P 500 and Nasdaq-100, gold, silver and oil. Every venue on this page is data on disk today; nothing here is a roadmap.
Per UTC day with an instrument or coin filter: pick a range, the tables you want and the instruments you want. Polymarket perps ride inside every Polymarket plan: Starter at BBO + trades, L10 and L25 in full. Hyperliquid is a plan bought for Hyperliquid: Starter ($29/mo, top of book and trades on every coin plus one coin's full book), L10 ($49/mo, every coin's book at 10 levels) or L25 ($99/mo, every coin's full 25-level book). For both venues' perps on one account, add Polymarket perps to a Hyperliquid plan for $29 a month, or hold a Polymarket plan next to it: each venue is its own subscription and they add up. Any account can pull the fixed sample day (2026-08-12) on both venues at full fidelity without spending a claim.
Parquet in a Hive layout (table/date=/symbol=/), so one read_parquet reads a whole table with date and symbol injected from the path, plus the instruments dictionary and a README that states the schemas and the coverage caveats for your exact range. Delivery is a signed download link or the REST API.
For spot and oracle price feeds alongside prediction-market books, see crypto tick data; for the prediction-market order book itself, see Polymarket historical data.