The agent successfully assembled a comprehensive explanation of Edge Network's business model, pricing, and differentiation by fetching the site's core product and pricing pages. The site proved highly navigable for this discovery task: it published a curated full-content artifact (llms-full.txt) and offered markdown versions of key pages (.md files), making structured information accessible without scraping fragmented HTML. The agent's final answer directly cited specifics from pricing pages, product descriptions, and the About page, with only minor ambiguities remaining around tokenization mechanics and decentralization implementation details.

• ›Step [2] (llms-full.txt) was critical: the site explicitly publishes a machine-readable full corpus for AI agents, labeled as such. This artifact contained the Agent API overview and anchored the agent's understanding of the platform's scope before it visited individual pages.
• ›Steps [4, 9] (pricing.md and cdn/pricing.md) provided exact pricing tiers, fee structures, and the zero-egress differentiator—the most concrete competitive claim. The .md versions were cleaner and more machine-readable than the HTML rendering of the same pages (steps [3, 8]).
• ›Steps [7, 11, 13, 16, 17] (About, Compute, Storage, Shield, Assist markdown pages) directly supplied the product portfolio, mission statement, and use-case positioning. The agent cited specific numbers and claims from these pages (e.g., '$0.004/vCPU-hour', 'no cookies, no fingerprinting', '1,000+ locations').
• ›The site's markdown artifact strategy (.md companion files alongside HTML) was agent-optimized: the agent discovered and preferred these over HTML, reducing parsing friction.
• ›Gaps identified by the agent (tokenization flow, node operator model, AWS/Cloudflare comparisons) were genuinely absent or vague in fetched content—not due to agent failure. The agent correctly flagged these as confusions rather than misunderstandings.
• ›The agent did not need to guess core facts (what they do, pricing, differentiation) because the site surfaced them directly in structured, citable pages. 76% of fetches came from previous_artifact (the llms-full.txt seed and .md files already known), reducing web-search dependency to 0%.