What L2 order-book data is, and why it matters for prediction markets

Most prediction-market datasets you can buy are trades: a record of fills — who bought or sold, at what price, when. Trades tell you what happened. They don't tell you what was available to trade at each moment, which is where a lot of the signal lives.

Level 1 vs Level 2 vs Level 3

Market data is usually described in levels of depth:

• Level 1 (top of book): the best bid and best ask, plus their sizes. The tightest quote at each moment.
• Level 2 (full depth): every resting order price level on both sides of the book, with the size available at each. This is the shape of supply and demand.
• Level 3 (order by order): each individual order, with add/cancel/modify events. The most granular, and rarely public on prediction venues.

SupaGamma captures L2 for prediction markets — the full book depth as it changed over time, reconstructed from the venue's own delta stream. On Polymarket and Kalshi the book is anonymized, so L2 is the realistic ceiling; on Hyperliquid, true order-by-order data is possible via a self-run node.

Why L2 is the perishable asset

Trades can often be backfilled — they're settled on-chain or in the venue's history. Order books usually cannot. Most venues only serve the current book over their API; once the book moves, the previous state is gone unless someone was capturing it continuously. That makes a gapless, timestamped L2 archive genuinely unbackfillable: if it wasn't recorded as it happened, it doesn't exist.

What you can do with it

• Measure real liquidity and depth, not just traded volume — how much you could have moved without walking the book.
• Reconstruct the spread and mid over time for microstructure research and fair-value work.
• Backtest execution: model slippage against the book that actually existed, instead of assuming trades filled at mid.
• Study how books reacted to news, resolution events, and cross-venue moves.

SupaGamma sells the raw L2 delta tape (the book-change stream as it happened) and reconstructed depth snapshots, in CSV, Parquet, and JSON, priced per MB with subscription options for heavier use.