How would you design a legally defensible data collecti...

The design should be layered: use third-party crawls for broad web data, but negotiate explicit licenses for high-value, high-risk content. You must actively exclude sources that prohibit AI training in their terms of service and respect robots.txt during any first-party crawling. Finally, maintaining per-document provenance logs is essential for responding to legal requests, such as data erasure.