Order books & data quality
Reconstruct observable state and understand what collection health can tell you.
On this page
Replay an L2 book
- Start from a snapshot
Collect every row belonging to the snapshot. Use the levels and sequence anchor actually supplied by the market.
- Apply subsequent updates
Follow the venue’s sequence rules. Replace each changed price-level quantity; zero quantity removes a level.
- Recover from gaps
Stop treating the book as continuous when the sequence check fails. Acquire a new valid anchor before resuming replay.
What a snapshot cannot recover
Repeated snapshots show observed book states. They cannot reveal individual orders, queue position, or intermediate changes that were never recorded. Clearing the book for every price-level row also loses the rest of the snapshot.
Exchange sequence IDs, checksums, and timestamps have different meanings. A universal sort by exchange timestamp is not a verified replay algorithm for every venue.
Health and completeness
| Signal | Tells you | Does not prove |
|---|---|---|
| Publication progress | The end of the latest complete UTC day confirmed by the publisher. | That every earlier historical partition is continuous or available for every instrument. |
| Sequence continuity | Whether adjacent updates follow the source rules. | That the source itself published every market event. |
| Coverage dates | The catalog’s declared historical window. | That every instrument traded throughout that window. |