Glossary
Not in the crawl
Material that never entered a public web crawl, so a model has not already seen it — worth more because it adds genuinely new signal.
Public scrapes such as Common Crawl feed effectively every foundation model, so material inside them adds little new signal. Private archives — unpublished masters, raw sessions, subscriber-only feeds, internal footage — sit outside that overlap. The claim is checked through provenance records and absence from public indexes.
Why it matters
For evaluation data the property is even stricter: only unexposed material yields a trustworthy benchmark.