The agent assembled a comprehensive explanation of GNU.org by working around direct site access blocks (403 errors on the main domain) and relying entirely on web search results and prior knowledge. It successfully answered what GNU does, who it's for, pricing, and how it differs from alternatives, but could not verify this understanding against the site's own hosted content—all answers came from search-surfaced pages and general knowledge rather than the navigable site itself.

• ›Direct fetch attempts to gnu.org returned 403 Forbidden errors (steps [1], [2], [6]), forcing the agent to pivot entirely to web search. The site was effectively non-navigable via automated HTTP requests.
• ›Web searches (steps [4], [7], [9], [10], [12], [13]) returned result snippets and linked pages (e.g., gnu.org/philosophy, gnu.org/software/software.html, gnu.org/licenses/gpl-faq.en.html) but the agent never successfully fetched any of these pages to verify their content. The agent's final answer was built from search result titles/URLs and prior knowledge, not from reading the actual pages.
• ›The agent correctly cited sources in its final response (gnu.org/philosophy, gnu.org/licenses/gpl-faq.en.html, gnu.org/software/emacs/, etc.), but these citations refer to pages it listed as search results, not pages it successfully retrieved and read. The citations are aspirational rather than grounded in fetched content.
• ›The final answer is accurate and well-structured, but derived from general knowledge of the GNU Project rather than from the site's own navigation. The agent acknowledged this limitation ('I couldn't directly access gnu.org's pages—they blocked my automated requests').
• ›The site's access restriction (blanket 403 responses) made it impossible for the agent to assess how well gnu.org itself surfaces information about mission, software offerings, licensing, and differentiation. The agent had to fall back on what it already knew and what search results disclosed.