A Root Snapshot is a field-only snapshot: a real, schema-conformant record that has field data but lacks the rich UI-derived artifacts (screenshots, stitched captures, reports) a full capture produces. It's made by an existing job type, not new machinery — the bot reads the imported record's opaque payload and writes each value straight into schema fields instead of crawling the UI. To stay current, that same payload-to-fields job is requeued nightly; it re-reads its source, checks for changes, and updates fields in place. A run can fan out to tens of thousands of jobs, but since each only refreshes data (no UI crawl), each finishes in seconds.

Why bother? AI search needs an indexed corpus to exist before any query can run, and the Root layer is that corpus during early Preserve — letting a user search millions of records immediately, with full Live Snapshots as the on-demand drill-down. It's optional, deployed only when a customer wants AI-search-enabled access early.