The agent assembled a coherent explanation of Referent's value proposition, target market, and pricing model, but relied heavily on prior knowledge (86%) rather than machine-readable content from the site itself. Direct fetches of referent.app returned truncated HTML that didn't expose structured information; the agent fell back on web search snippets and inferred positioning from the tagline 'AI for the systems that have no connector.' The site is navigable for human visitors but hostile to agent extraction—no pricing page, no docs, no API reference, and key details (exact pricing, security certifications, onboarding timelines) remain inaccessible.

• ›Fetches [1, 3, 5] returned only truncated HTML boilerplate (language-redirect script); no main content was exposed in the response bodies, forcing the agent to rely on prior knowledge and search results rather than live site data.
• ›Web searches [11–19] returned title tags and meta descriptions ('AI for the systems that have no connector,' 'One price, the whole app, yours forever') that matched the agent's final summary, but the actual site content was not mechanically extracted—the agent synthesized positioning from search result metadata.
• ›The site publishes no dedicated pages for features (/features → 404), about (/about → 404), how-it-works (/how-it-works → 404), or comparisons (/vs → 404), leaving the agent to infer differentiators (vs. RPA, vs. Moveworks) from external sources rather than first-party positioning.
• ›Pricing is custom-quote-only with no public tiers or numbers published—the agent correctly reported this limitation but could not cite a pricing page because none exists in a machine-readable form.
• ›Security, compliance, and technical constraints (data encryption, storage location, SOC 2 status, rate limits, on-premises support) were documented as 'could not find' in the final response, confirming the site lacks technical/compliance documentation layers.