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 panel.