The agent successfully assembled a comprehensive explanation of CORRECTIV by combining website exploration with web search and external sources. The site itself provided limited English-language detail about operations, funding, and differentiation, forcing the agent to rely heavily on Wikipedia, Reuters Institute case studies, and third-party profiles to answer the full scope of the task (what they do, who it's for, pricing, differentiation). The website was navigable but sparse in English; most substantive content required prior knowledge or external search.

• ›Step [13] (FAQ), [14] (Finances), and [15] (main investigations page) fetched successfully but returned HTML structure without visible rendered content in the trajectory—the agent drew on these pages plus prior knowledge and web search results to construct statements about funding model (~40% donations, ~30% grants, ~30% economic activity) and programs (CORRECTIV.Europe, CORRECTIV.Lokal, Exile).
• ›Steps [17] (CORRECTIV.Europe) and [20] (Wikipedia) were explicitly cited as sources and likely contained substantive material about the international network and founding context, respectively, but the HTML was truncated in the trajectory, indicating the agent synthesized information across multiple fragments.
• ›Early attempts to fetch /en/about and /en/services returned 404s, signaling poor English URL structure. The agent worked around this by guessing /en/support, /en/investigations, /en/europe, /en/finances, and /en/about-us/faq—some succeeded, some failed. This suggests the site's English information architecture is incomplete or inconsistently documented.
• ›The agent relied on web search (steps [10], [11], [16], [18], [19]) to find Reuters Institute case study, Wikipedia entry, and CredCatalog profile—none of these are hosted on correctiv.org itself. The final answer cites 9 external sources vs. only 3 pages from correctiv.org's own domain, indicating the site does not self-contain the contextual information needed to explain its mission, differentiation, and impact.
• ›The agent correctly identified and called out confusion points (limited English content, unclear service distinctions, vague external offerings) as gaps in the site's documentation, showing honest assessment rather than hallucination fill-in.