Claimthread
Status: Curated research demonstration · experimental AI unverified
My role: Product direction, reading experience, acceptance criteria and AI-assisted design and implementation
Technology: Astro, TypeScript and a separate React research interface
The problem
A confident claim can hide a lot of assumptions. Claimthread explores how readers could follow the reasoning behind a claim without losing their place whenever they open a source.
The curated demonstration examines claims about AI and water. It separates national estimates, local scarcity and the assumptions behind an almond-versus-ChatGPT comparison.
My decisions and the workflow
I directed an editorial reading experience that breaks the comparison into four chapters: the reported claim, its inputs, an accounting assumption and the conclusion. Source records open alongside the reading flow and return readers to the same chapter.
I kept qualifications in the main explanation. For example, the almond footprint includes a pollution-dilution indicator, which is not additional water physically consumed. The demonstration reconstructs the arithmetic without turning it into a universal environmental verdict.
Implementation and demonstrated result
Astro renders the curated pages and source records. Small scripts support the chapters and charts; a separate React interface contains the experimental research workflow. AI tools assisted design, implementation and verification under my product direction.
The September 21 handoff recorded 16 passing tests, a clean Astro check and a production build. Desktop and mobile walkthroughs covered all four chapter-to-source returns and keyboard focus. These checks show interface behavior, not improved comprehension or research accuracy.
Limits and next step
Live AI Gateway operation remains unverified. A matching quotation can establish where text came from without proving that a conclusion follows from it. The portfolio therefore presents curated content and screenshots, with no public AI-demo claim.
The next useful evaluation is whether readers can identify an assumption, locate its source and explain what the evidence does—and does not—support.