General FAQs

What is DeepTick?

DeepTick is a real-time normalization engine and historical database for high-frequency cryptocurrency market data. It handles WebSocket connection state, reconnects, snapshot seeding, and data serialization into uniform Parquet/CSV formats.

Where is the data hosted?

All normalized datasets are archived and stored as hourly/daily partitions inside globally replicated AWS S3 storage buckets. Clients can download partitions directly using our HTTP API or native SDKs.

Do you support real-time feeds?

Yes. In addition to historical partitions, DeepTick offers direct WebSocket streaming feeds providing normalized real-time updates for trades, liquidations, and order book deltas.

What timestamp precision do you use?

All timestamps are microsecond-precision UTC stored as int64. Every record has both timestamp (exchange-reported) and local_timestamp (our receipt time), enabling accurate latency analysis.

Which exchanges support options or L3 data?

The schemas support options summary chains, options trades, and L3 order lifecycle fields. Published coverage depends on the exchange connector and validation status; options IV/Greeks fields are populated only where the exchange feed or collector provides them.

How do I get free sample data?

See the Free Data section on our homepage. Free-tier access includes published sandbox datasets for selected assets and first-of-month dates.

Data & Licensing

What is the delay on historical datasets?

The target product shape is daily verified partitions, with fresher intra-day files only when the collector and compaction jobs have published them. Current availability should be checked against the catalog/status page rather than assumed for every exchange.

Can I use the datasets for commercial purposes?

Yes, all DeepTick datasets downloaded with a valid API key support commercial backtesting, quantitative research, and production ingestion pipelines. Redissemination of raw data is prohibited.

What's the difference between Parquet and CSV downloads?

Parquet is up to 8x smaller, supports column pruning (download only columns you need), and loads directly into Pandas, Polars, or Arrow without string parsing. CSV is human-readable but substantially larger.

How do I download data for many dates at once?

Use the Client SDK's CachedClient.load_trades() (or equivalent language methods) with a date range parameter. It automatically parallelizes partition downloads and caches files locally to prevent redundant downloads.

Order Books

How are order book deltas represented?

Deltas are represented as arrays of price-size pairs. If a price level's size is 0, it indicates the level has been depleted and should be deleted from your reconstructed book.

How do I build a local order book snapshot from deltas?

Load the raw book_l2_delta partitions and use the derived-data utilities to reconstruct BBO or snapshots locally. The public SDK wrapper around full book rebuilding is still being hardened, so treat exchange-specific sequence handling as part of validation.