A Data-Quality Contract for Crypto Backtests
Pin venue, timestamps, units, gaps, and lineage before trusting simulated results.
Key takeaways
- Pin market and schema.
- Audit temporal integrity.
- Preserve immutable raw lineage.
A Data-Quality Contract for Crypto Backtests: the decision context
Identify venue, instrument, quote, market type, multiplier, and symbol history. Validate samples against specifications before normalizing identifiers.
Event, receipt, and bar timestamps answer different questions. Document fields, timezone, ordering, duplicates, and bar construction.
What evidence deserves attention
Gaps, outages, outliers, and zero-volume bars need explicit treatment. Produce an anomaly report and preserve raw alongside transformed data.
A repeatable review workflow
Audit one symbol-day by checking timestamp monotonicity, duplicate keys, interval gaps, high-low consistency, nonnegative volume, quote/base units, and a sample against raw trades. Save the source URI, retrieval time, checksum, timezone, schema version, and every transformation used to create bars.
Limits, failure modes, and risk
Vendor history can be revised. Store retrieval time, checksum, schema, and conversion code while acknowledging that clean data does not prove liquidity.
Passing these checks cannot reveal all bad ticks, delisted-symbol changes, hidden outages, or whether historical quotes were executable. Vendor backfills can rewrite history, and synthetic candles omit queue position, latency, and intrabar ordering needed for many strategies.
Frequently asked questions
What is the first thing to distinguish in A Data-Quality Contract for Crypto Backtests?
Start with this article's central checkpoint: Pin market and schema. Then verify the definition and scope against the cited sources.
How can I check A Data-Quality Contract for Crypto Backtests in practice?
Use the worked procedure in the article and keep these two checks together: Audit temporal integrity. Preserve immutable raw lineage.
What is the most important limitation?
Data defects can manufacture apparent performance, and clean history still cannot guarantee execution.
How this article was prepared
This educational article was prepared with AI assistance, then reviewed editorially for clarity and checked against the cited source material.