The agent successfully assembled a comprehensive understanding of Raft Computer by fetching the homepage, about page, service pages, sitemap, and an LLM-optimized summary (llms.txt). However, critical information gaps remain unfilled—pricing is entirely opaque, case studies are absent, and the site is deliberately minimalist with only 4 public pages. The agent had to bridge these gaps with prior knowledge and web search results for competing products, rather than finding answers on the site itself.

• ›The agent's main content came from four direct fetches: the English homepage [3], about page [10], two service pages [37, 38], and the LLM-specific summary [39]. Together these provided a coherent positioning statement (who they serve, how they work, continuity claim), but almost no operational detail.
• ›Raft Computer explicitly surfaces machine-readable context via robots.txt and llms.txt—the latter being a text summary written for LLM consumption. This is unusually agent-friendly, yet the site's actual page content is heavily truncated in the responses, suggesting either aggressive HTML minimization or client-side rendering that the fetcher could not fully render.
• ›The site publishes zero pricing information, no case studies, no FAQ, no blog, and no contact form—all routes returned 404. The agent had to rely on the assessment-based sales model being described in prose rather than any transactional interface. This is intentional minimalism, not oversight.
• ›Web search steps [14–28] were routers only—they returned results for other products (Sourcegraph Amp, raft.build, raft.ai) but not for raft.computer itself, confirming the domain has minimal web visibility outside its own site. The agent correctly did not cite these.
• ›The sitemap [34] was the critical discovery tool—it revealed which pages exist (only 4 public URLs). Without it, the agent would have remained stuck after 404s on /pricing, /services, /blog, /faq, /contact, and /how-it-works.