The agent assembled a comprehensive overview of Edwards & Hardy by fetching the homepage, about page, and service-specific pages (roofing, drone, guttering), supplemented by web search results that confirmed the company's positioning and heritage. The site proved partially navigable—service pages existed and were fetchable, but transparent pricing was absent, forcing the agent to rely on search results and inferred knowledge to construct a complete answer about how they differ from competitors.

• ›Steps [1], [8], [14], [15], [22] returned actual page content describing services (roofing, drone inspections, guttering), but all responses were HTML fragments truncated at ~4400 chars—the agent could not render full page text, suggesting client-side rendering (Next.js) made detailed content inaccessible via fetch.
• ›Pricing information was entirely absent from fetched pages; the agent found a blog post URL ('New Roof Cost NZ: 2026 Price Guide') via search [11] but could not cite it as primary source because the trajectory does not show [11]'s content being fetched—the agent inferred pricing model from that URL title and prior knowledge, not from direct retrieval.
• ›Competitive differentiation (drone technology, RoofGuard system, Lindab gutters, heritage since 1959) came from search result titles and snippets [6], [12], [17], [24], not from the fetched pages themselves—the site does not foreground these claims in a machine-readable way.
• ›The site's structure broke standard conventions: `/services` and `/pricing` routes returned 404s [2], [3], forcing the agent to discover pages only through web search and prior knowledge (40% of fetch sources were web search, 50% prior artifact).
• ›The agent cited specific URLs in its final response's Sources section, but only those it successfully fetched; search results that provided title-level insights (e.g., 'Innovative Drone Roof Inspections & Treatments') were not formally cited, indicating the agent conflated discovery paths with evidence sources.