Grade every [S#] claim in an answer against the exact evidence snippet it cites, and flag any claim the source does not actually support. Use whenever an agent produces medical or scientific claims tagged with numbered sources.
