Polymarket Order Book History

• Sports books: a consistent 35–36 price levels per side on connected grids — the deepest rest in the study.
• Up/Down books: a ladder that compresses near strike resolution and expands after it; the same series alternates between a full ladder and a single level.
• The archive records full depth as-of each frame; a market's history is replayable second to second (as-of queries), not just summarizable.
• One-sidedness is a constant: at least 90% of sampled frames carry size on one side, so full-depth history is most often a one-leg book.

The 2026-09-09 ETH repricing is a full-depth as-of story: the ladder's executable floor re-priced to the futures level at 14:42:50, and the mid to 0.065, in the same second the depth axes swung.

Depth directionality is measurable: on the August 23 event, ETH bid depth fell 96% while ask depth grew 491% — a full-ladder asymmetry invisible to top-of-book archives.

Final-minute frames are the hardest: ask presence in 3 of 18 sampled one-second frames means the full-ladder record is the only honest description of exit liquidity.

• GET a market's books at 250ms between two timestamps; every record is a full-depth frame with sequence, ask/bid levels and sizes.
• Reconstruct as-of: index frames by ts and query the last frame at or before a moment — the point-in-time replay primitive.
• Change series with one parameter; the wording of levels (0.50/0.60 ... 0.99) and depth (bid ask of each level, sizes) is the same schema for every market.

• Pull one sports market's books at 250ms for an hour and count distinct price levels per side per frame; sports should hold 35–36 levels per side on connected grids.
• Pull the same hour for an Up/Down market and count levels again; the ladder that shrinks to a single level near contract end is a documented shape.
• Replay the last second before a documented reprice as-of: index by timestamp, take the last frame at or before the move, and describe its depth in both directions.
• Compute the one-sided ratio per side and print it next to the average level count — the two numbers together describe liquidity shape, not just size.

A full-ladder archive is the difference between knowing the top of book and knowing the book; the depth swings of a repricing (ETH bid −96%, ask +491%) are invisible to top-of-book feeds entirely.

The as-of replay capability is the archive's signature: any past second, reconstructed exactly, is the reproduction primitive behind every settlement and reprice claim.

One-sidedness means most of your history-reading will be single-leg; the full-depth frame is the only record that tells you honestly when the other leg exists.

Book history feeds replays, settlement studies, and market-structure dashboards; format it Parquet with (market, ts) sort and it merges with every other series trivially.

Because books, prices, and metrics share the timestamp grid, the book archive is your foundation, not a silo.

The book archive is the reconstruction layer: full-depth frames at 250ms make any past second replayable exactly, and the as-of query is the primitive underneath every settlement and repricing study published here.

Sports books at 35–36 levels per side and Up/Down ladders that compress to a single level are two book anatomies in one record — the same schema for completely different liquidity shapes.

One-sidedness is your data-quality cab: when 90%+ of frames rest a single side, single-side depth is the norm, and any top-of-book-only feed by construction cannot tell you the half of it.

Keep the record as Parquet sorted on (market, ts) alongside prices and metrics; three series, one grid, one foundation for everything downstream.

• order book archive
• order book snapshots
• order book replay data