The agent gathered partial information about Shuffled by combining site metadata (from the homepage and sitemap), the archive page structure, and comparisons to documented alternatives found via search. However, the heavily JavaScript-rendered frontend prevented access to full page content, visual design, and feature details. The site's minimal structure (no about, pricing, or FAQ pages) and lack of online coverage forced the agent to assemble answers from fragments and prior knowledge rather than from clearly published resources.

• ›Step [1] (homepage fetch) returned only the HTML skeleton with Next.js bundle references; the actual page content is client-rendered and invisible to static HTML fetches. The agent relied on meta tags and schema markup to extract 'three daily puzzles' and '$0 USD pricing' — the site's only machine-readable positioning.
• ›Step [5] (sitemap.xml) revealed only two URL patterns: the homepage and /archive, confirming the site has no dedicated about, pricing, or FAQ pages. This forced the agent to infer positioning rather than read it directly.
• ›Step [7] (archive page fetch) returned the same JS-only skeleton, providing no visibility into past puzzle content or how the archive works — another feature the agent could not verify from the HTML.
• ›Steps [14] and [17] (Crosswordle and Swapdle competitor sites) were content-bearing for the comparison table. The agent explicitly cited these in its Sources section and used them to position Shuffled relative to alternatives, but Shuffled's own site provided almost no differentiation details.
• ›The site is intentionally minimal and game-focused, but this design choice makes it agent-hostile: no discoverable text content, no FAQ or help docs, no creator/team information, and no online press coverage or third-party articles that might fill the gaps. The agent had to rely almost entirely on metadata inference and competitor research.