Binance historical order book data: snapshots, updates and coverage

By Cruscy Data · 2026-10-30

Historical Binance order-book data describes the bids and asks that were recorded at particular times. It answers a different question from a candle: what liquidity was visible, rather than only where trades occurred during a bar.

Choose the market before choosing the file format

BTC/USDT spot and the BTCUSDT USD-M perpetual have separate order books, trade streams and execution rules. A spot snapshot is not a substitute for a futures snapshot, even when the prices are close. Keep the venue, market and symbol attached to every research result.

In Cruscy Data, l2_BTCUSDT describes spot book updates. The futures equivalent is fl2_BTCUSDT. The f prefix is a market distinction, not a higher-quality version of the same book.

Snapshots and updates solve different problems

A snapshot supplies a reference state. Incremental updates describe changes after that state. Reconstructing a book requires the right starting snapshot, ordered updates, and continuity checks using the exchange update identifiers. Sorting records by arrival time alone does not repair a missing update.

A fixed-depth product is easier to inspect when your model needs only nearby liquidity. Our spot top-25 book records 25 levels per side once per second. The separate full update stream supports research that needs to follow changes between those observations. These products have different sampling rules; compare their schemas before joining them.

Check coverage you can actually query

A provider's first collection date, its archived dates and the dates readable under your plan can differ. Check the stream's current coverage, market and quality report. A saved archive does not by itself prove continuous recording or immediate SQL access to every historical day.

Trades and candles cannot restore the exact book updates that were never recorded. If a session has a gap, retain that gap in your experiment and disclose how it affects the usable window. A later reference snapshot can restart a reconstruction, but it does not recover the missing historical changes.

A useful first comparison

Choose one symbol and one short, documented window. Compare the reconstructed best bid and ask with an independently sampled book product, inspect any sequence breaks, and check the published quality results. Then decide whether you need raw updates, fixed-depth snapshots or just top-of-book features. The field dictionary defines our timestamps, scales and update identifiers.