How to Export Polymarket Order Books

Each row is a snapshot: a UTC timestamp and the full bid/ask ladder — every level's price and size, bids sorted best-first, asks best-first. Add the /prices and /metrics endpoints on the same grid and a single market export becomes a complete microstructure dataset.

Resolved markets export with the winning outcome on every row, which is what makes an export a study instead of a souvenir.

Pull the final 5 minutes of a resolved BTC 5-minute market at 250ms: ~1,200 snapshots that include the empty-side minutes (76.2% of final-minute books) and the mid-42k rest at the bound.

Exported, that window becomes the exact material for the fill simulation and replay recipes on this site.

An export is only as good as its grid: coarser resolution merges frames and hides single-sidedness, so the honest default is 250ms for anything settlement-adjacent, and coarser grids should be labeled as views, not as the raw record.

CSV is universally readable but bloats at 250ms scale; JSON keeps the nesty bid/ask structure intact one-to-one; Parquet is where the size stops mattering. The pragmatic path is Parquet for analysis, JSON for API-shaped consumption, CSV for anything that must open in a spreadsheet.

Whichever you pick, keep the timestamp grid explicit. A full-L2 export is only as useful as the clock it carries, and the archive's 250ms UTC buckets are the join key for every other series you will pull.

• Order book data
• Order book history
• Book API guide