Crypto liquidity and order book depth: a historical data audit

What went wrong in the archive

The underlying research sampled BTC, SOL and LINK book-depth archives from January 2023 through November 2025, comparing matched weekdays within each month. A first completeness screen retained 309 of 315 asset-days. A later quality audit found 12 additional asset-days with implausibly static bid or ask depth, leaving 297. Completeness alone had not caught the problem.

Some affected days repeated identical values throughout the day and across separate dates. Other days had a frozen bid side while asks continued changing. A file can therefore contain many timestamps without containing many genuine market updates. The older interpretation based on the contaminated sample was withdrawn.

Coverage, variation and units are separate checks

The audit checks key uniqueness, hour coverage, valid positive depth and price-band availability, then inspects each side for static patterns and cross-date repetition. A highly one-sided book is a review flag, not automatic proof of corruption: genuine stress can also produce imbalance. If a matched event or comparison day is invalid, the whole asset-month comparison is excluded.

The available historical bands measure bid depth within 1% below and ask depth within 1% above the reference price. They do not measure the narrow spread or depth closest to execution. A result about a 1% band cannot establish what happened to tighter bands or to the fill price of a specific order.

What survives the correction

For the later comparison period, 17 valid matched months remain. Full-day geometric bid-depth ratios are 1.037 for BTC, 1.076 for SOL and 1.028 for LINK. These are within-sample ratios against the matched reference days, not percentage shares of market liquidity. They do not support a general statement that month-end bids were thinner in the measured band.

SOL retains an unadjusted positive bid-depth difference, but its ask side also rises. That is not evidence of a bid-only improvement. Multiple comparisons were not adjusted. Failing to find thinner bids does not prove equal liquidity, and these observations cannot be extrapolated to all assets, spot markets or every month-end date.

Volume does not identify the cause of a price move

Executed volume and resting depth describe different parts of the market. Lower turnover need not mean fewer bids, and a price decline can itself change reported volume and depth. To argue that liquidity caused a move, the relevant deterioration should precede the price change and survive controls for volatility, time of day and asset selection.

Data revisions belong in the research record. Publishing the retained sample, exclusion categories and corrected conclusion is more informative than retaining a stronger story supported by invalid observations. This article reports the existing corrected audit; it does not claim a new raw-file download or an independent replication of every snapshot.

What does crypto order book depth measure?

Depth describes displayed buying or selling interest within a stated price band on a specified venue. It is different from trading volume, and it does not guarantee the execution price or available size of a future order.

Does a complete historical order book file prove good liquidity data?

No. This audit found additional static-data anomalies after an initial completeness check. Repeated timestamps with frozen values can create an appearance of coverage without genuine market updates.

Limitations

Three representative assets, selected weekdays and coarse depth bands limit generalisation. Monthly observations across assets are correlated. This exploratory comparison does not identify order flow, participant intent or a causal calendar effect.

Sources and data access

• Binance public data and archive formats

Cited sources document external facts or data formats; they do not endorse our findings. The public snapshot contains the table and methods. Source-file fingerprints are included for recalculated datasets. Full archived raw files are not distributed here; fingerprints identify the files used, rather than independently proving the results.

Aggregate data snapshot (JSON) · Markdown · Methods and limitations

Live prior 30-day hourly baseline — separate from the historical study

Only the current watch list is shown, up to 12 assets. Missing or stale estimates remain unavailable. Venue and quote currency are kept separate; beta needs matching BTC data. Hourly statistics are not minute-move probabilities.

Observation cutoff (UTC): 20/10/2026, 14:58

Related research