The agent successfully assembled a comprehensive explanation of Cartwright by combining multiple content sources: the homepage (step [9]), documentation introduction (step [8]), llms.txt file (step [12]), and the docs index (step [6]). The site's information was scattered across different formats and required the agent to synthesize fragments rather than publish a single unified overview, but all core claims (what it does, who it's for, pricing, differentiation) were grounded in retrieved content. The demo.cartwright.app itself was inaccessible due to initial timeouts, but the main cartwright.app domain provided sufficient material.

• ›Step [9] (homepage markdown) and step [8] (docs/introduction) provided the core positioning: 'AI-native commerce engine' that you own rather than rent, with MCP/agent-native architecture. These were the semantic bedrock for the entire answer.
• ›Step [12] (llms.txt) was crucial for pricing structure and technical stack details ('MIT-licensed', 'Next.js 16 app with Prisma, Stripe, Auth.js') that don't appear verbatim on the main marketing pages—the agent had to discover this artifact by guessing a well-known file path.
• ›Step [6] (docs index page) was primarily a router; step [3] (homepage) returned HTML that the agent did not parse deeply. The agent relied on markdown artifacts (steps [9], [8], [12]) rather than rendered HTML, suggesting these are better-structured for machine consumption.
• ›The site publishes intent well via markdown files and structured docs, but lacks a single 'comparison matrix' or 'pricing & positioning' page that directly answers all four questions (what, who, price, vs. alternatives) in one place. The agent had to cross-reference multiple pages.
• ›Demo.cartwright.app was inaccessible (timeouts on steps [0], [1]; eventually loaded as HTML on step [4], [11] but agent did not extract text from it). The agent fell back to www.cartwright.app entirely, showing the main site is the authoritative source for understanding the product.
• ›The agent correctly identified gaps: 'AI operates the shop' is not concretely exemplified on the marketing site; enterprise tier pricing is opaque; the demo store is read-only for visitors. These are legitimate surface-area friction points, not agent navigation failures.