[ns-ui] A per-sentence grounding trace under an AI answer: solid where a source backs the claim, a bare hairline where the model is inferring alone, diagonal hatch where a source contradicts it. Shape-encoded, never color, with a static legend keying each mark to its meaning and a click-to-raise source pane
pnpm dlx shadcn@latest add "https://registry.directory/r/nikolas-sapa/citation-grounding-hatch.json"