BTC · 5 minutes
The raw parquet folded down to one row per second: open, high, low and close of the mid price and of the trades, volume, the best bid and ask and the top ten levels of the book on each side. Built from the raw files of the same day, and every file records the sha256 of the one it came from.
- Available
- 2026-03-24 → 2026-10-09
- Days on sale
- 200
- Size
- 135 GB
- Per day
- 690 MB
- Rows
- 20.4B
- Format
- parquet, one file per source per UTC day
What one day contains
The Polymarket 5 minutes up-down market for BTC, plus the exchange data for the same coin on the same clock. The exchange files are included in the price — they belong to the coin, not to the market, so whichever of the two markets you buy, they come with it.
- polymarket.btc.5m — the market itself
- binance.btc — Binance
- bybit.btc — Bybit
- okx.btc — OKX
- coinbase.btc — Coinbase
- chainlink.btc — Chainlink
What the data looks like
One row per second per instrument — for Polymarket, per market outcome — on the collector's
receive clock. A row stamped ts covers [ts, ts + 1 s).
ts · bid_price · bid_size · ask_price · ask_size mid_open · mid_high · mid_low · mid_close · mid_twap open · high · low · close · vwap · volume · trades bid_price_00 … bid_price_09 · bid_size_00 … bid_size_09 ask_price_00 … ask_price_09 · ask_size_00 … ask_size_09 n_messages · book_valid · book_age_ms
The footer of every file names the raw file it was built from and its sha256, the bucket rule and the schema version, so an aggregate can be checked against its source.
Coverage
Share of the UTC day in which the file holds at least one row, by receive time. It says how much of the day arrived — not how much of it was worth trading: an illiquid pair is quiet because nobody traded, and Chainlink publishes once a second or two by design, so it sits near 92% on a perfectly good day.