Three layers, all recorded event-by-event rather than sampled. The Hyperliquid order books and fills for every coin since 2025-01-25 — top of book and the 25-level book on every change, every trade with the taker's wallet — rebuilt from the chain's own order-event stream. The crypto price feeds — exchange spot, the Chainlink oracle and the 30s/60s TWAPs — at roughly ~2.5M rows a day. And the full order-book history of every Polymarket crypto market trading against them.
Parquet, one read_parquet per table. No feed to run, no gaps to backfill, no reconstruction after the fact — and no $10,000 S3 egress bill for the chain archive.
A full day of a real Polymarket BTC market, not a synthetic excerpt: L1 quotes on every change, 25-level L2 depth on every book update, and every trade, with the reference dictionary and a worked pandas notebook alongside. Parquet and CSV in the same download. No account, no card, no form. Read the schema before you decide anything. With a free account there is a second sample: a full Hyperliquid day (2026-08-12) for any coins you like, up to 1 GB, books and fills included — pull it from the Hyperliquid catalog.
Past the sample, a free account claims 5 more market·days from July 2026.
Coverage dates are capture-start dates, and capture has been continuous since.
Every crypto prediction market on the venue, captured tick-by-tick: BTC, ETH, SOL, XRP, DOGE, BNB and HYPE up-or-down at 5m and 15m, the hourly multi-strike ladders, and the longer-dated price targets. L1 quotes on every change, 25-level L2 depth on every book update, and the trade tape.
The price stream those markets settle against, recorded tick-by-tick from Polymarket's own real-time data service: exchange spot, the Chainlink oracle, and the 30s/60s TWAPs that actually resolve the up/down markets. Sold as whole days.
Hyperliquid's order books rebuilt from the chain's own order-event stream: top of book on every change, the 25-level book on every change with resting-order counts per level, and every fill with the taker's wallet, fee and realized PnL — for every coin the venue has listed since January 2025 (BTC, ETH, SOL, HYPE, the spot pairs, the HIP-3 builder dexes), continued daily. Sold per day with a coin filter; see /hyperliquid-historical-data.
Polymarket's perpetual futures venue, captured tick-by-tick: BTC, ETH, SOL, XRP, DOGE, BNB, HYPE and the rest of its crypto perps (24 of its 67 instruments; the others are US equities, indices and commodities). Event-driven BBO, 20-level books every ~100 ms, the full ladder every ~15 s, every trade, and tickers with mark, index, open interest and funding. Sold per day with an instrument filter.
Every price tick Polymarket's own real-time data stream publishes: exchange spot, Chainlink oracle, and the 30s/60s TWAPs that actually resolve the crypto up/down markets. This is the reference series a crypto market settles against. About ~2.5M rows and 150–200 MB per day.
Exchange spot prices. Symbol form is <asset>usdt.
Oracle prices. Symbol form is <asset>/usd. Carries hype and zec, which spot does not.
One of the two topics that RESOLVE Polymarket crypto up/down markets. Days of history, not weeks.
The second market-resolving TWAP topic. Days of history, not weeks.
Most people who go looking for one are not trying to build market-data infrastructure. They want a clean history of what the book did, and writing a feed handler is the tax standing in the way: a socket that reconnects, sequence numbers that must be checked, snapshots reconciled against deltas, a storage layout that survives a year of it, and the discovery that none of it works retroactively. Whatever you did not capture yesterday is gone.
We have run that stack continuously since 2026-05-11 on dual-region, sequence-audited collectors, and normalized the output into Hive-partitioned parquet. What arrives is a dataframe.
If you genuinely do want the wire format — an auditor rebuilding a book at a specific nanosecond, or a desk that will not take anyone else's normalization on trust — the raw layer ships the untouched websocket messages, compressed and hash-audited, with the deterministic replay engine. That is the Custom tier.
A row per event, not a row per interval. Quotes are written when the best bid or ask changes, book updates when the ladder changes, trades when a fill prints, and price-feed ticks when the upstream publishes one. Nothing is sampled on a timer or bucketed into candles before it reaches you. Polymarket rows carry the collector's nanosecond receive timestamp; Hyperliquid rows carry the chain's block time and height, one row per block in which the book changed.
Yes, and it is the deepest set on the site: every perp, spot pair and HIP-3 market since 2025-01-25, reconstructed from the chain's public order-event archive rather than recorded from the websocket (which only publishes throttled snapshots). Top of book and the 25-level book on every change, and every fill with taker wallet, fee and PnL. Per day with a coin filter; a free account can pull the fixed sample day. See /hyperliquid-historical-data.
Yes, and it needs no account. The sample is a full day of a real Polymarket BTC market — L1 quotes, 25-level L2 depth and the trade tape, in parquet with a CSV twin, plus the reference dictionary and a worked pandas notebook. Download it from the samples page and read it before you decide anything.
Exchange spot carries btcusdt, ethusdt, solusdt, xrpusdt, dogeusdt and bnbusdt. The Chainlink oracle and both TWAP topics carry btc/usd, eth/usd, sol/usd, xrp/usd, doge/usd, bnb/usd, hype/usd and zec/usd — so the oracle side additionally covers HYPE and ZEC, which spot does not.
No. That is the point of buying this rather than building it. We have run dual-region, sequence-audited collectors continuously since May 2026 and normalized the output into Hive-partitioned parquet, so what you get is a dataframe, not a socket to babysit. If you do want the untouched websocket messages plus the deterministic replay engine, that is the raw layer on the Custom tier.
Parquet with a CSV twin in the same bundle, plus a reference dictionary mapping every id back to the real market. One read_parquet gives you a dataframe. Delivery is a signed download link, the REST API, or SFTP.
For the whole archive rather than the crypto slice, see Polymarket historical data; for perpetual futures books, funding and open interest, see perps tick data, or browse every crypto market on disk.