Downloads

v2.0

Published 2026-08-23. The 2011 MtGox series manifest.json, SHA256SUMS

mtgox

MtGox, BTC/USD (bankrupt since 2014), quoted in USD, 177,585 rows, 2011-09-21 to 2011-10-20. MtGox best bid and ask, their midpoint and spread, plus two distance-weighted depth scores. bidscore and askscore sum book volume within $0.50 of the best price on their side, weighted by 1/(1+|price-best|)^2 and floored, and diff is their log ratio. THAT IS A THIRD DEPTH DEFINITION, unrelated to the 2014 top-60 ratio and to the modern +/-2% band: three eras, three definitions, never one series. Prices come from single-precision float(6,4) storage, so they carry four decimals. There is no trade price in this series, only the book midpoint.

v1.0

Published 2026-08-23. The 2014-2015 six-venue series manifest.json, SHA256SUMS

bayesdata_bitfinex

Bitfinex, api.bitfinex.com/v1 btcusd, quoted in USD, 10,123,356 rows, 2014-12-17 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 0.45% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 100 (the source integer divided by this is the price, so two decimals of precision).

bayesdata_bitstamp

Bitstamp, bitstamp.net/api btcusd, quoted in USD, 3,390,136 rows, 2014-12-17 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 1.79% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 100 (the source integer divided by this is the price, so two decimals of precision).

bayesdata_btce

BTC-e, btc-e.com/api/2 btc_usd (seized by the FBI in 2017), quoted in USD, 2,244,786 rows, 2014-12-17 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 0.02% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 1000 (the source integer divided by this is the price, so three decimals of precision).

bayesdata_okcoin

OKCoin, okcoin.com/api USD book, quoted in USD, 9,185,180 rows, 2014-11-02 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 0.05% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 100 (the source integer divided by this is the price, so two decimals of precision). The canonical dump contributed 8468499 rows from 2014-12-17 20:22:00. The wyvern fragment then contributed 716681 rows read, of which 716681 were new and 0 were dropped as datelines the canonical dump already held, extending coverage back to 2014-11-02 11:49:01.

bayesdata_btcchina

BTCChina, data.btcchina.com btccny (closed in 2017), quoted in CNY, 10,838,969 rows, 2014-12-17 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 0.03% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 100 (the source integer divided by this is the price, so two decimals of precision).

bayesdata_okcoin_cny

OKCoin CNY, okcoin.cn/api/v1 btc_cny (died with the PBoC ban), quoted in CNY, 2,124,649 rows, 2014-12-23 to 2015-11-07. Top-of-book plus a book-imbalance ratio, 2 s nominal cadence drifting with fetch time. THE 2014 book_ratio IS NOT THE MODERN ONE: it is (bid_volume - ask_volume) / (bid_volume + ask_volume) over the top 60 levels per side of whatever the venue returned, not a fixed price band, and must not be joined to any other era. Zero ratios are kept as collected because a failed depth fetch and a balanced book are indistinguishable row by row. Placeholder contamination ceiling for this table: 14.69% of its rows carry a zero ratio, which is an upper bound because a genuinely balanced book stores zero too. Source scale: 100 (the source integer divided by this is the price, so two decimals of precision).

What is not here

Two kinds of thing are published: the closed historical eras of market data, and the fee index, which is derived every week from public on-chain aggregates rather than collected. The series this project has been collecting since 2024 is neither, and it is not published: not the Bitstamp and Kraken books, not Ether, Solana or Litecoin, not Bitfinex or MEXC, and nothing cross-venue. Its methodology is written down because the older files were collected by the same hand, but there are no files and no date. That is a standing policy, enforced in the exporter rather than in a paragraph, and only the operator changes it.