Goooddy/crypto-lob-stream

Multi-Exchange L2 Order Book, Trades & Futures Data

Continuous Level-2 order book (depth diffs), trades, periodic full-book snapshots, and (for perpetual futures venues) funding rate, liquidation, and open interest data, captured with crypto-lob-stream and published monthly as a contribution to open market-microstructure research.

Full depth-of-book data is paywalled or licensed for most asset classes; crypto is the exception, where the raw feeds are genuinely public. This dataset exists to lower that data barrier for independent researchers and students -- now across eight exchanges (spot and perpetual futures) rather than one.

Coverage

Exchange	Assets	Data from
binance	BTCUSDT, ETHUSDT, SOLUSDT (from 2026-06-01); XRPUSDT, DOGEUSDT, LINKUSDT, AVAXUSDT, DOTUSDT, LTCUSDT, ADAUSDT (from 2026-09-01)	—
binance_futures	BTCUSDT, ETHUSDT, SOLUSDT	2026-09-01
coinbase	BTC-USD, ETH-USD, SOL-USD, XRP-USD, DOGE-USD	2026-09-01
okx	BTC-USDT, ETH-USDT, SOL-USDT, XRP-USDT, DOGE-USDT	2026-09-01
okx_swap	BTC-USDT, ETH-USDT, SOL-USDT	2026-09-01
kraken	BTC/USD, ETH/USD, SOL/USD, XRP/USD, DOGE/USD	2026-09-01
bybit	BTCUSDT, ETHUSDT, SOLUSDT, LINKUSDT, ADAUSDT	2026-09-01
bybit_linear	BTCUSDT, ETHUSDT, SOLUSDT	2026-09-01

Layout

Each table is partitioned {prefix}/{exchange}/{asset}/YYYY-MM.parquet. Files are Snappy-compressed Parquet, one file per month per exchange/asset, in crypto-lob-stream's native schema (including the exchange column, and exchange_ts alongside timestamp_ms on trades/depth/funding/ liquidations/open_interest as of package version 0.9.0), so its own reconstruction helper works directly against this dataset with no conversion step.

Backward compatibility

This dataset covered Binance only before this release. If your code already loads depth, trades, or snapshots with no exchange suffix, it keeps working unchanged -- those names still point at Binance's data specifically (now covering more pairs than before, as Binance's own coverage has grown). Every exchange, including Binance, is also available under its explicit {prefix}_{exchange} name (e.g. depth_binance, depth_okx, depth_kraken) for anyone who wants a specific venue.

Reconstructing the order book

Depth rows are diffs, not a standing book. Replay a snapshot plus every subsequent diff, pruning to your intended depth after each update, to rebuild the book at any instant.

Heads up: a naive replay (only dropping a level when quantity hits 0) accumulates "ghost levels" over time -- price levels that fell out of scope but were never explicitly zeroed out, so they just sit there looking real. There's a nasty failure mode worth knowing about: a replay can be completely gap-free and still end up wrong.

Oliver Zehentleitner ran an independent 25.10-hour BTCUSDT test on this exact problem (gap-free, sequence-validated, no synthetic gaps at all) found that a naive, unpruned cache grew to 20,758 bid levels and 9,116 ask levels. At the final audit against REST, only 24.09% of those bid levels and 39.82% of the ask levels still matched. A pruned cache, checked the same way, held steady at 1,011 bids / 1,078 asks with 87.83% / 91.74% matching REST. So don't skip the pruning step -- sequence continuity alone doesn't stop stale levels from piling up.

The crypto-lob-stream package handles this for you and prunes automatically:

pip install crypto-lob-stream

from crypto_lob_stream import reconstruct

book = reconstruct("./local_copy_of_this_dataset", exchange="binance", asset="BTCUSDT")
bids, asks = book.top(n=10)   # correctly pruned

If you're not using Python, the manual version is: load the nearest snapshot before your target window, discard diffs at or before its last_update_id, apply the rest in ascending last_update_id order (each diff's quantity is the new total, not a delta -- 0.0 means remove), and prune back to your intended depth after every single update. That last step is the one that actually prevents the ghost-level problem above.

See the package README for exchange-specific default pruning depths and further detail.

Known limitations

Binance depth and trades data for BTCUSDT/ETHUSDT/SOLUSDT from June through August 2026 was collected by this project's original, standalone v1 pipeline, which had a flush-naming bug causing later flushes within the same clock hour to silently overwrite earlier ones (see the v0.9.2 release notes for the mechanism). That pipeline was retired without the fix ever being applied to it, so this affects its entire operating history, not a partial window -- the current pipeline (crypto-lob-stream 0.9.2+) has never had this issue, and every other exchange's data collection began on it directly.

Real, measured impact per month, from an audit of every published depth/trades file for this period:

Month	Hours affected	Estimated volume retained
June 2026	~8%	~93%
July 2026	~14%	~88%
August 2026	~52%	~52%

August specifically lost roughly half its expected depth/trades volume for these three pairs -- worth treating as significantly degraded for that month, not a minor gap. snapshots was never vulnerable to this issue in any month.

Full explanation: see the v0.9.2 release notes. Exact affected hours, file by file: see limitations_section.md.

Known gaps

July 2026: capture paused 2026-07-05 20:56 to ~21:39 UTC (43 min) for Binance BTCUSDT/ETHUSDT/SOLUSDT, across depth and trades, due to a host restart. A fresh snapshot was written on reconnect, so books before and after the gap reconstruct cleanly -- treat the gap as a hard reset and do not replay a diff sequence across it. (This is unrelated to the flush-naming issue above -- see the Known limitations section for that one specifically.)

Where this comes from

Collected with crypto-lob-stream (GitHub, MIT licensed, same as this dataset), a package for streaming L2 order book, trade, funding, liquidation, and open-interest data off public exchange WebSocket feeds straight to Parquet. Issues and PRs welcome on the repo.