Run /graphify on a document-heavy corpus (markdown reports, CSV data, JSON scan
fragments) that is ABOUT a fixed set of real-world entities (services, systems,
investigation objectives, technologies) without the graph fragmenting into
per-file ghost duplicates.
