The agent assembled a comprehensive explanation of OrcaRouter by combining fragments from the JavaScript-heavy homepage, pricing page, blog post, and web search results—none of which fully loaded their interactive UI. The site's lack of accessible documentation pages (404s on /docs, /features, /about) forced heavy reliance on web search and prior knowledge (64% sourced from prior knowledge) to answer questions about features, differentiation, and target customers. Despite this friction, the agent delivered a credible answer by cross-referencing external sources (academic paper, press releases, GitHub) and the site's own blog comparison post.

• ›Steps [1], [3], and [8] returned HTML with stylesheet references but truncated content, indicating the site is heavily JavaScript-rendered and not agent-friendly; the agent could not extract full page content from homepage, pricing, or blog fetches.
• ›Search results (steps [14], [15], [21], [22]) were more content-bearing than direct site fetches—they returned meta-descriptions, article titles, and URL hints that guided the agent to specific pages (like the OpenRouter comparison blog); the search snippets themselves contained positioning clues ('zero-markup', 'adaptive routing').
• ›Step [17] (blog comparison post) was cited in sources but also returned truncated HTML, meaning the agent inferred the comparison table from search result titles and external knowledge rather than from readable page content.
• ›The site lacks structured data (no JSON-LD, no schema.org markup for pricing or features) and no FAQ, docs, or about page—discovery and identity layers are undermined by document unavailability and JavaScript-dependent rendering.
• ›The agent successfully pivoted to web search and academic/press sources to fill gaps, suggesting the site's public relations and external coverage are more accessible than the website itself.